Distributed new energy air conditioner collaborative optimization method and system based on energy consumption portrait

Abstract

The invention discloses a distributed new energy air conditioner collaborative optimization method and system based on an energy consumption portrait. The method comprises the steps that historical user energy consumption related data is collected; feature extraction is carried out based on the collected data, and user energy consumption feature vectors of the preset areas are generated; according to a preset portrait label generation rule, generating a portrait label library based on the user energy consumption feature vector of each region; constructing a collaborative optimization model; the collaborative optimization model comprises a first scheduling layer and a second scheduling layer; before each optimization period, inputting the portrait label library and the electricity price into a first scheduling layer, and generating a distributed new energy air conditioner collaborative optimization instruction in the optimization period; and operating the distributed new energy air conditioner collaborative optimization instruction, and adjusting the distributed new energy air conditioner collaborative optimization instruction based on the second scheduling layer in real time. The method has the outstanding advantages that a complete portrait driving control closed loop is formed, collaborative optimization can be carried out on the equipment layer, and a collaborative optimization instruction giving consideration to balance of the comfort demand and the power-saving demand is generated.

Description

Technical Field

[0001] This invention relates to the field of new energy technology, specifically to a method and system for collaborative optimization of distributed new energy air conditioning based on energy consumption profiles. Background Technology

[0002] Traditional energy management systems often suffer from problems such as rudimentary operation, sacrifice of user comfort, and lack of user profiling. Most systems rely on simple electricity pricing strategies or fixed start-stop logic, failing to consider users' actual energy consumption habits, resulting in low photovoltaic absorption rates and poor energy storage charging and discharging efficiency. In pursuit of energy conservation, they forcibly adjust air conditioning temperatures, leading to a poor user experience and even user complaints. They only monitor basic electricity consumption and lack the ability to model user energy consumption behavior in multiple dimensions, making it impossible to generate personalized strategies.

[0003] Currently, none of the methods for generating personalized strategies based on energy consumption profiles have formed a complete profile-driven control closed loop.

[0004] An existing method for generating personalized strategies based on energy consumption profiles includes: Step S1: Acquiring basic information data, electricity consumption behavior data, and electricity consumption feedback data of an enterprise, and preprocessing them to obtain an enterprise electricity dataset; Step S2: Mining frequent itemsets and association rules from the enterprise electricity dataset using an association rule mining algorithm, and merging them with the enterprise electricity dataset to obtain a final feature dataset; Step S3: Based on the final feature dataset, using a clustering algorithm to construct a fine-grained user profile of the power enterprise; Step S4: Constructing an enterprise electricity cost model; Step S5: Based on an intelligent recommendation algorithm, combining the user profile and the enterprise electricity cost model, recommending the most suitable energy solution for the enterprise; Step S6: Integrating the intelligent recommendation algorithm into the enterprise energy management system. However, this method is not suitable for distributed energy management scenarios and does not consider users' sensitivity to electricity prices or collaborative optimization at the equipment layer.

[0005] Another existing method for generating personalized strategies based on energy consumption profiles includes: Step 1: Constructing an optimized configuration of integrated energy technologies for industrial parks that considers multi-energy coupling relationships; Step 2: Establishing a tag-based integrated energy system with multi-entity collaborative low-carbon economic operation technology; Step 3: Constructing a comprehensive demand response collaborative optimization model that considers cross-category energy. However, this method mainly addresses the economic issues of energy service projects for industrial enterprises and integrated energy service providers. It fails to solve the problem of balancing comfort needs with energy-saving needs. Summary of the Invention

[0006] To address the shortcomings of existing technologies, such as the disconnect between energy management system operation and actual user energy demand, the neglect of individual comfort preferences in air conditioning adjustment strategies, and the lack of adaptive capabilities to dynamically respond to changes in user behavior, this invention provides a distributed new energy air conditioning collaborative optimization method and system based on energy consumption profiles.

[0007] The present invention adopts the following technical solution.

[0008] The first aspect of this invention discloses a method for collaborative optimization of distributed new energy air conditioning based on energy consumption profiles, comprising: Collect historical user energy consumption data; Based on the historical user energy consumption data, feature extraction is performed to generate user energy consumption feature vectors for each preset area; Based on the preset profile tag generation rules, a profile tag library is generated based on the user energy consumption feature vectors of each region. A collaborative optimization model is constructed; the collaborative optimization model includes a first scheduling layer for generating distributed new energy air conditioning collaborative optimization instructions within a preset optimization period, and a second scheduling layer for adjusting the collaborative optimization instructions in real time; Before each optimization cycle, the profile tag library and electricity price are input into the first scheduling layer to generate the distributed new energy air conditioning collaborative optimization instruction for that optimization cycle. The distributed new energy air conditioning collaborative optimization command is executed, and the room temperature in each area is collected in real time and input into the second scheduling layer to adjust the distributed new energy air conditioning collaborative optimization command.

[0009] More preferably, The historical user energy consumption data includes the electricity consumption of each device in each region, the duration of stay in the region, and the number of people staying in the region.

[0010] More preferably, Each user energy consumption feature vector for each region includes spatial weighting coefficients, time period labels, temperature tolerance, and price sensitivity.

[0011] More preferably, The step of extracting features based on the historical user energy consumption data to generate user energy consumption feature vectors for each preset region includes: Based on the historical user energy consumption data, feature extraction is performed to obtain the spatial weight coefficient, time period label, temperature tolerance, and price sensitivity for each region. A user energy consumption feature vector for each region is constructed based on the spatial weight coefficient, the time period label, the temperature tolerance, and the price sensitivity.

[0012] More preferably, The step of extracting features based on the historical user energy consumption data to obtain the spatial weight coefficient, time period label, temperature tolerance, and price sensitivity for each region includes: The spatial weight coefficient of each region is calculated based on the region area, the length of stay in the region, and the number of people staying in the region. Based on a preset clustering algorithm, the historical electricity consumption data of each region are clustered to determine the high-frequency energy consumption periods of each region, and a time period label is generated for each region based on the high-frequency energy consumption period. The temperature tolerance of a region is calculated based on the magnitude and number of times the user adjusts the temperature in each region within a preset first time period. The price sensitivity of a region is calculated based on the electricity price and the corresponding electricity consumption in that region.

[0013] More preferably, The step of generating a profile tag library based on the user energy consumption feature vectors of each region according to preset profile tag generation rules includes: According to the preset profile tag generation rules, spatial activity tags, energy-active time period tags, temperature preference tags, and electricity preference tags are generated for each region based on the user energy consumption feature vector of each region. Based on the spatial activity tags of each region, the energy-active time period tags, the temperature preference tags, and the electricity preference tags, a corresponding profile tag set is generated; Based on the image tag sets of each region, an image tag library is constructed.

[0014] More preferably, The profile tag generation rules include sub-rules for generating spatial activity tags, sub-rules for generating energy-active time period tags, sub-rules for generating temperature preference tags, and sub-rules for generating electricity preference tags; The spatial activity tag generation sub-rule compares the spatial weight coefficient of each region with a preset first spatial weight threshold and a second spatial weight threshold to determine the spatial activity level of each region; based on the preset region name and the corresponding spatial activity level of the region, the spatial activity tag is generated accordingly; the first spatial weight threshold is less than the second spatial weight threshold. The sub-rule for generating active energy consumption period labels is as follows: It calculates the active period weight based on the average electricity consumption of the cluster corresponding to the high-frequency energy consumption period in each region and a preset baseline electricity consumption; it compares the price sensitivity of each region with preset first price sensitivity thresholds and second price sensitivity thresholds to determine the price sensitivity level of each region; and it generates the active energy consumption period label based on the active period weight, period label, and price sensitivity level of each region; the first price sensitivity threshold is greater than the second price sensitivity threshold. The temperature preference label generation sub-rule compares the temperature tolerance of each region with a preset first temperature tolerance threshold and a second temperature tolerance threshold to determine the temperature preference label for each region accordingly; the first temperature tolerance threshold is less than the second temperature tolerance threshold. The sub-rule for generating electricity preference tags is based on a preset correspondence between price sensitivity level and electricity preference, and generates the electricity preference tag according to the price sensitivity level of each region.

[0015] More preferably, After generating the image tag library, it is determined whether the image tag set of each region needs to be self-evolved based on the preset first image self-evolution rule, second image self-evolution rule and third image self-evolution rule respectively. When it is determined, based on the first self-evolution rule of the profile, that the profile tag set of a certain region needs to be self-evolved, the temperature preference tag in the profile tag set of the region is updated to generate a self-evolved profile tag set; the first self-evolution rule of the profile is that when the frequency of user temperature adjustment in a certain region is greater than the set adjustment frequency threshold within a set second time length, it is determined that the profile tag set of the region needs to be self-evolved. When it is determined, based on the second self-evolution rule of the portrait, that the portrait tag set of a certain region needs to be self-evolved, the spatial activity tag in the portrait tag set of the region is updated to generate a self-evolved portrait tag set; the second self-evolution rule of the portrait is that when the rate of change of spatial weight coefficient in a certain region within a set third time length is greater than the set threshold of the rate of change of weight coefficient, it is determined that the portrait tag set of the region needs to be self-evolved. When it is determined, based on the third profile self-evolution rule, that the profile tag set of a certain region needs to be self-evolved, the energy-active period tag and the electricity preference tag in the profile tag set of that region are updated to generate a self-evolved profile tag set; the third profile self-evolution rule is that when the change in outdoor temperature within a set fourth time length is greater than a set outdoor temperature change threshold, it is determined that the profile tag set of all regions needs to be self-evolved. If it is not determined based on the first image self-evolution rule, the second image self-evolution rule, and the third image self-evolution rule that the image tag set of any region needs to undergo self-evolution, then it continues to determine whether the image tag set of each region needs to undergo self-evolution based on the first image self-evolution rule, the second image self-evolution rule, and the third image self-evolution rule.

[0016] More preferably, The first scheduling layer includes a collaborative optimization objective function and collaborative optimization constraints; The collaborative optimization objective function aims to minimize the cost-comfort fusion data. The collaborative optimization constraints include energy storage operation constraints, air conditioning power constraints, air conditioning temperature setting constraints, and power balance constraints.

[0017] More preferably, The cost-comfort fusion data is calculated based on the amount of electricity purchased by the user from the power grid, the electricity price, the comfort penalty weighting coefficient, the air conditioning set temperature of each region, and the user's comfort preference temperature in each region. The comfort penalty weight coefficient is determined based on the energy-saving preference label corresponding to the region and the preset correspondence between the energy-saving preference label and the comfort penalty weight coefficient. The user comfort preference temperature for the area is calculated based on the user's preset baseline value within the area and the temperature offset obtained based on the preset correspondence between energy-active period labels and temperature offsets.

[0018] More preferably, The second scheduling layer calculates a power adjustment amount based on the room temperature and the air conditioner set temperature in a certain area when the absolute value of the deviation between the room temperature and the air conditioner set temperature in that area is greater than the temperature tolerance. The frequency adjustment amount of the air conditioner compressor is determined based on the power adjustment amount; The frequency of the air conditioner compressor is adjusted based on the aforementioned air conditioner compressor frequency adjustment amount.

[0019] More preferably, When the power grid issues a demand response command, the average power purchased by the user from the power grid within the set fifth time period is collected as the first power purchased, and the area to be adjusted is selected from each of the preset area selection rules. Based on the outdoor temperature and the room temperature of each of the areas to be adjusted, the maximum adjustable power of each of the areas to be adjusted is calculated accordingly. Based on the price sensitivity, dwell time, and number of residents in each of the regions to be adjusted, the regional adjustment priority of each of the regions to be adjusted is calculated. The region to be adjusted corresponding to the maximum value among the adjustment priorities of each region is taken as the target adjustment region; the power of the target adjustment region is adjusted based on the maximum adjustable power of the target adjustment region; After adjusting the power in the target adjustment area, the average power purchased by the user from the grid within the set fifth time period is collected as the second power purchase. When the deviation between the first purchased power and the second purchased power is greater than or equal to the set power threshold, and there is still the area to be adjusted, the area to be adjusted corresponding to the maximum value among the adjustment priorities of each area is taken as the target adjustment area; based on the maximum adjustable power of the target adjustment area, the power of the target adjustment area is adjusted until the deviation between the first purchased power and the second purchased power is less than the set power threshold, or there is no area to be adjusted.

[0020] More preferably, The region selection rule is that when the electricity price sensitivity of a certain region is higher than a preset third price sensitivity threshold and the spatial activity is lower than a preset activity threshold, the region is selected as the region to be adjusted; the spatial activity of each region is calculated based on the region's dwell time and the number of people residing in the region.

[0021] The second aspect of this invention discloses a distributed new energy air conditioning collaborative optimization system based on a distributed new energy air conditioning collaborative optimization method, including a user energy consumption data acquisition module, a user energy consumption feature vector generation module, a profile tag library generation module, a collaborative optimization model construction module, a collaborative optimization instruction generation module, and a collaborative optimization instruction real-time adjustment module. The user energy consumption data acquisition module is used to collect historical user energy consumption data; The user energy consumption feature vector generation module performs feature extraction based on the historical user energy consumption related data to generate user energy consumption feature vectors for each preset area. The image tag library generation module generates an image tag library based on the user energy consumption feature vector of each region according to the preset image tag generation rules. The collaborative optimization model construction module is used to construct a collaborative optimization model; the collaborative optimization model includes a first scheduling layer for generating distributed new energy air conditioning collaborative optimization instructions within a preset optimization period, and a second scheduling layer for adjusting the collaborative optimization instructions in real time. The collaborative optimization instruction generation module is used to input the profile tag library and electricity price into the first scheduling layer before each optimization cycle to generate the distributed new energy air conditioning collaborative optimization instruction for that optimization cycle. The real-time adjustment module for the collaborative optimization instruction is used to run the distributed new energy air conditioning collaborative optimization instruction, and to collect the room temperature input of each area in real time into the second scheduling layer to adjust the distributed new energy air conditioning collaborative optimization instruction.

[0022] The third aspect of the present invention discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the distributed new energy air conditioning collaborative optimization method.

[0023] The beneficial effects of this invention are compared with those of the prior art: 1. This invention constructs a multi-dimensional energy consumption profile tag library model based on non-intrusive load monitoring technology, integrating four-dimensional data: time, space, environment (comfort dimension), and behavior (economic dimension). By constructing a multi-dimensional energy consumption profile tag library, the system can accurately identify users' behavioral habits and environmental preferences, and generate personalized control strategies to improve user comfort and system energy efficiency.

[0024] 2. This invention uses a collaborative optimization model driven by a profile tag library, built on a two-layer optimization control logic, to coordinate the optimization of energy storage charging and discharging plans and air conditioning operation strategies. This can improve the local photovoltaic absorption rate and energy storage utilization efficiency, thereby achieving the technical effect of improving overall energy utilization.

[0025] 3. This invention utilizes a self-evolutionary profile mechanism, enabling the system to dynamically adjust extracted feature parameters based on user feedback and environmental changes. This allows for continuous optimization of distributed new energy air conditioning collaborative optimization instructions, thereby enhancing the system's intelligence level. 4. The present invention has outstanding advantages such as forming a complete image tag library-driven control closed loop, being able to perform collaborative optimization of various devices, and generating collaborative optimization instructions that balance comfort and energy saving requirements. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the process of the distributed new energy air conditioning collaborative optimization method based on energy consumption profile of the present invention; Figure 2 This is a schematic diagram of the overall architecture of a distributed new energy air conditioning collaborative optimization system based on energy consumption profiles. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0028] The first aspect of this application discloses a collaborative optimization method for distributed new energy air conditioning based on energy consumption profiles, see appendix. Figure 1 ,include: Historical user energy consumption data is collected; this data includes the electricity consumption of each device in each region, the duration of stay in the region, and the number of people staying in the region. Preferably, the electricity consumption data is calculated every minute; those skilled in the art should know how to calculate electricity consumption based on power, and this will not be elaborated here.

[0029] The present invention collects historical user energy consumption data, which is a preparatory step for extracting user energy consumption feature vectors. Based on this data, it is possible to understand the operating status of equipment in each area and time period, as well as the frequency of user use in each area, making the extracted features more representative. This enables the present invention to generate collaborative optimization instructions that balance comfort and energy saving requirements. Based on the historical user energy consumption data, feature extraction is performed to generate user energy consumption feature vectors for each preset region; each user energy consumption feature vector for each region includes spatial weight coefficients, time period labels, temperature tolerance, and price sensitivity.

[0030] The user energy consumption feature vector adopted in this invention comprehensively considers four dimensions: time, space, comfort, and economy. It reflects users' energy consumption habits from various aspects, thereby enabling a better understanding of users' comfort and energy-saving needs in different areas. This makes the profile tags of each area recorded in the profile tag library generated by this invention more accurate, laying a solid foundation for generating collaborative optimization instructions that balance comfort and energy-saving needs.

[0031] The step of extracting features based on the historical user energy consumption data to generate user energy consumption feature vectors for each preset region includes: Based on the historical user energy consumption data, feature extraction is performed to obtain the spatial weight coefficient, time period label, temperature tolerance, and price sensitivity for each area in the building floor plan, including: The spatial weight coefficient W for each region is calculated based on its area, duration of stay, and number of residents. s The formula is as follows: W s = (Area of ​​the region × Spatial activity level of the region) ÷ Σ(Area of ​​all regions × Spatial activity level of the corresponding region); Wherein, Σ(area of ​​all regions × spatial activity of the corresponding region) means that after multiplying the area of ​​each region by its corresponding spatial activity, all product results are summed.

[0032] The spatial activity level of each area is calculated based on the duration of stay and the number of people residing in that area, as shown in the following formula: Spatial activity level of a certain area = total person-hours in that area ÷ sum of total person-hours in all areas; The formula for calculating the total per capita hours in the region is as follows: Total person-hours in the region = (Duration of stay in the area during the tth stay × Number of people in the area during the tth stay) Where t represents the t-th time a person stays in the area within the set first time period; T represents the total number of times a person stays in the area within the set first time period; when the area occupancy status of a certain area remains unchanged, but the number of people residing in the area changes, the area residence time is recalculated, and the number of times a person stays in the area is incremented by one; for example, if the area occupancy status of a certain area remains at 1, but the number of people residing in the area changes from 2 to 1, the number of times a person stays within the set first time period changes from 3 to 4. The preferred value range for the set first time period is 1 day; The historical electricity consumption data of each region are clustered based on a preset clustering algorithm to determine the high-frequency energy consumption periods of each region, and a time period label is generated for each region based on the high-frequency energy consumption period; preferably, the preset clustering algorithm is a density-based noisy application spatial clustering algorithm (DBSCAN clustering algorithm for short). Specifically, based on a preset clustering algorithm, the historical electricity consumption data of each region are clustered to determine the high-frequency energy consumption periods of each region, including: Based on historical electricity consumption data for each region, a time-electricity consumption correlation matrix is ​​constructed. The rows and columns of this matrix represent different initial time periods obtained by dividing 24 hours, and the columns and rows represent electricity consumption. In this matrix, the intersection of rows and columns records the average electricity consumption for the corresponding time period from the historical electricity consumption data used. Preferably, the different initial time periods obtained by dividing 24 hours are different initial time periods obtained by dividing 24 hours by an average, and the length of each initial time period is preferably in the range of 5 to 30 minutes. Preferably, one month's worth of historical electricity consumption data is used.

[0033] Based on a preset clustering algorithm, the historical electricity consumption data of each region are clustered to obtain multiple clusters; When the minimum power consumption of each initial period in the cluster is greater than or equal to the preset peak period baseline level, or greater than or equal to the first baseline level gain coefficient × the normal period baseline level, the period corresponding to the cluster is taken as the high-frequency energy consumption period; the preferred value range of the first baseline level gain coefficient is 1.1~1.3, and this coefficient can be adjusted by those skilled in the art to ensure that at least one high-frequency energy consumption period is obtained.

[0034] The peak-hour baseline level and the normal-hour baseline level were obtained based on the following methods: The time-electricity consumption correlation matrix is ​​clustered and identified using the k-means algorithm to obtain three clusters. The average electricity consumption of each of the three clusters is calculated, which is the cluster center. The maximum average electricity consumption is used as the baseline level during the peak period, and the median average electricity consumption is used as the baseline level during the normal period. Since multiple high-frequency energy consumption periods may be obtained, when generating period labels, those skilled in the art can select according to the actual situation. For example, a high-frequency energy consumption period can be randomly selected to generate a period label, or the period with the highest average electricity consumption among the high-frequency energy consumption periods can be selected to generate a period label.

[0035] The generation of time period tags for each region based on its high-frequency energy consumption periods is determined by the location of each region's high-frequency energy consumption periods within the day and the time range / average electricity consumption of the cluster corresponding to those periods. Specifically, the generation of time period tags for each region based on its high-frequency energy consumption periods includes: Based on the location of high-frequency energy consumption periods within a day, a first sub-label for each period is constructed. Further, when the high-frequency energy consumption period falls between 0:00 and 6:00 (inclusive of 0:00, exclusive of 6:00), the first sub-label is "Night"; when it falls between 6:00 and 12:00 (inclusive of 6:00, exclusive of 12:00), the first sub-label is "Morning"; when it falls between 12:00 and 18:00 (inclusive of 12:00, exclusive of 18:00), the first sub-label is "Afternoon"; and when it falls between 18:00 and 24:00 (inclusive of 18:00, exclusive of 24:00), the first sub-label is "Evening". Based on the time range of the cluster corresponding to this time period, or the average electricity consumption, a second sub-label for the time period is constructed, including: Determine whether the time range of the cluster corresponding to the time period is less than or equal to a set time period determination threshold; the preferred value of the time period determination threshold is 30 minutes. If the time range of the cluster corresponding to the time period is less than or equal to the set time period determination threshold, then the second sub-label of the time period is a transition; If the time range of the cluster corresponding to the time period is greater than the set time period determination threshold, then it is further determined whether the average power consumption of the cluster corresponding to the time period is greater than or equal to the product of the second baseline level gain coefficient and the peak time period baseline water; the preferred value range of the second baseline level gain coefficient is 1.2~1.5; the second baseline level gain coefficient is greater than or equal to the first baseline level gain coefficient.

[0036] If so, then the second sub-label of the time period is the core; If not, the second sub-label of the time period is empty.

[0037] When the second sub-label is not empty, the time period label is constructed according to the method of time period first sub-label + peak + time period second sub-label; for example, when the time period first sub-label is evening and the time period second sub-label is core, the time period label is evening peak core.

[0038] When the second sub-label is empty, the time period label is constructed according to the method of the first sub-label of the time period + the inter-peak segment; for example, when the first sub-label of the time period is evening and the second sub-label of the time period is empty, the time period label is the evening peak segment.

[0039] The time period labels exist in the form of numerical codes, for example, defining the evening peak core as 1, the morning peak core as 2, etc.

[0040] The temperature tolerance of a region is calculated based on the magnitude and number of times the user adjusts the temperature in each region within a preset first time period, as shown in the following formula: ΔT = Tolerance gain coefficient × Sum of temperature adjustments made by the user in this area within the first time period ÷ Number of temperature adjustments made by the user in this area within the first time period; Wherein, ΔT represents the temperature tolerance. The range of temperature adjustment by the user in this area is the absolute value of the difference between the temperature adjustment by the user in this area and the user's preset reference value. Preferably, the tolerance gain coefficient is 0.5, the first time length is 1 day, and when the user does not make a preset, the user's preset reference value is 26℃. The price sensitivity of a region is calculated based on the electricity price and the corresponding electricity consumption in each region; the price sensitivity of a region is calculated based on the electricity price and the electricity consumption in the first and second electricity consumption periods within each region. Based on the principles that the first and second electricity consumption periods are adjacent, the electricity prices in the first and second periods are different, and the electricity prices remain unchanged within the first and second periods, those skilled in the art should know how to select the first and second electricity consumption periods; the formula for calculating the price sensitivity is as follows: κ = [ (ΔQ / Q) / (ΔP / P) ] × e -0.05t ; Where κ represents price sensitivity (0≤κ≤1), e -0.05t The time decay factor is used, ΔQ / Q represents the relative rate of change in electricity consumption, ΔQ refers to the change in user electricity consumption between the first and second time periods, Q refers to the benchmark electricity consumption, ΔP / P represents the relative rate of change in electricity price, ΔP refers to the change in electricity price between the first and second time periods; P refers to the benchmark electricity price, e -0.05tThe time decay factor is represented by t, which represents the interval between the user's most recent electricity price-sensitive behavior and the current time, in hours. 0.05 is the decay coefficient, in units of 1 / h. Those skilled in the art should know how to obtain the benchmark electricity consumption and benchmark electricity price, so they will not be elaborated here.

[0041] This invention proposes a method for extracting the above-mentioned features, which improves the implementation effect of this invention.

[0042] The user energy consumption feature vector for each region is constructed based on the spatial weight coefficient, time period label, temperature tolerance, and price sensitivity of each region. That is, the spatial weight coefficient, time period label, temperature tolerance, and price sensitivity of each region are combined into a vector form as the user energy consumption feature vector of that region.

[0043] The user's energy consumption feature vector F is shown in the following formula: F = [spatial weighting coefficient, time period label, temperature tolerance, price sensitivity].

[0044] This invention generates user energy feature vectors from the four-dimensional features of each region, effectively reducing classification errors caused by the confusion of features in each region, thereby improving the overall reliability of the invention.

[0045] According to preset profile tag generation rules, a profile tag library is generated based on the user energy consumption feature vectors of each region, including: According to the preset profile tag generation rules, spatial activity tags, energy-active time period tags, temperature preference tags, and electricity preference tags are generated for each region based on the user energy consumption feature vector of each region. The image tag generation rules include sub-rules for generating spatial activity tags, sub-rules for generating energy-active time period tags, sub-rules for generating temperature preference tags, and sub-rules for generating electricity preference tags.

[0046] The spatial activity tag generation sub-rule compares the spatial weight coefficient of each region with a preset first spatial weight threshold and a second spatial weight threshold to determine the spatial activity level of each region. Based on preset region names (such as kitchen, living room, etc.) and the corresponding spatial activity level of the region, a spatial activity tag is generated. The first spatial weight threshold is less than the second spatial weight threshold. Specifically, when the spatial weight coefficient of a region is less than or equal to the first spatial weight threshold, the spatial activity level of that region is determined to be low. Preferably, the first spatial weight threshold is 0.5. When the spatial weight coefficient of a certain region is less than or equal to the second spatial weight threshold and greater than the first spatial weight threshold, the spatial activity level of the region is determined to be medium active; preferably, the second spatial weight threshold is 0.8. When the spatial weight coefficient of a certain region is greater than the second spatial weight threshold, the spatial activity level of that region is determined to be high activity.

[0047] The process of generating a spatial activity label based on preset area names (such as kitchen, living room, etc.) and the corresponding spatial activity level of the area refers to merging the area name and the corresponding spatial activity level of the area, adding the word "area", and a spatial weight coefficient as the spatial activity label.

[0048] The sub-rule for generating active energy consumption period labels is based on the average electricity consumption of the cluster corresponding to the high-frequency energy consumption period in each region and a preset baseline electricity consumption to calculate the active period weight. The baseline electricity consumption is preferably the average of all historical electricity consumption data in the corresponding region. The price sensitivity of each region is compared with a preset first price sensitivity threshold and a second price sensitivity threshold to determine the price sensitivity level of each region. Based on the active period weight, period label, and price sensitivity level of each region, the active energy consumption period label is generated accordingly. The first price sensitivity threshold is greater than the second price sensitivity threshold. The price sensitivity of each region is compared with a preset first price sensitivity threshold and a second price sensitivity threshold to determine the price sensitivity level of each region, including: When κ ≥ a first price sensitivity threshold, the price sensitivity level is high; preferably, the first price sensitivity threshold is 0.7. When the second price sensitivity threshold ≤ κ < the first price sensitivity threshold, the price sensitivity level is medium sensitive; preferably, the second price sensitivity threshold is 0.4. When κ < the second price sensitivity threshold, the price sensitivity level is low.

[0049] The process of generating an energy-consuming active time period label based on the active time period weight, time period label, and price sensitivity level of each region refers to merging the first character of the time period label and the price sensitivity level of the region, and adding the active time period weight as the energy-consuming active time period label. The temperature preference label generation sub-rule compares the temperature tolerance of each region with a preset first temperature tolerance threshold and a second temperature tolerance threshold to determine the temperature preference of each region and generate the temperature preference label; the first temperature tolerance threshold is less than the second temperature tolerance threshold. The step of comparing the temperature tolerance of each region with a preset first temperature tolerance threshold and a second temperature tolerance threshold to determine the temperature preference of each region includes: When ΔT < the first temperature tolerance threshold, the temperature preference of this region is strictly isothermal; preferably, the first temperature tolerance threshold is 0.3℃. When the first temperature tolerance threshold ≤ ΔT < the second temperature tolerance threshold, the temperature preference of this region is mild and constant; preferably, the second temperature tolerance threshold is 0.8℃. When ΔT ≥ the second temperature tolerance threshold, the temperature preference in this region is relaxed and comfortable.

[0050] Temperature preference labels are generated based on temperature preferences and temperature tolerance.

[0051] The sub-rule for generating electricity preference tags is based on a preset correspondence between price sensitivity levels and electricity preferences, generating electricity preference tags according to the price sensitivity level of each region. The correspondence between price sensitivity levels and electricity preference tags includes: High sensitivity (first price sensitivity threshold ≥ 0.7) corresponds to the electricity preference label "energy saving preference"; Medium sensitivity (second price sensitivity threshold ≤ κ < first price sensitivity threshold), corresponding to the electricity preference label "economic equilibrium type"; Low sensitivity (κ < second price sensitivity threshold) corresponds to the electricity preference label "comfort-first".

[0052] Electricity preference labels are generated based on the price sensitivity level and price sensitivity of each region. This involves determining electricity preferences based on the price sensitivity level and price sensitivity of each region, and generating electricity preference labels based on these preferences and price sensitivity. For example, when κ=0.7, the price sensitivity level is high sensitivity, the corresponding electricity preference is energy-saving preference, and the generated node preference label is "energy-saving preference (κ=0.7)". A profile tag set is generated based on the spatial activity tag, energy-active period tag, temperature preference tag, and electricity preference tag of each region.

[0053] This invention provides a specific method for setting profile labels for each dimension, enabling the invention to be implemented more effectively. By setting profile labels for each dimension according to the method of this invention, the collaborative optimization model can better perform collaborative optimization based on the profile labels.

[0054] Based on the spatial activity tags of each region, the energy-active time period tags, the temperature preference tags, and the electricity preference tags, a corresponding profile tag set is generated; Based on the image tag sets of each region, an image tag library is constructed.

[0055] This invention constructs a multi-dimensional energy consumption profile tag library based on non-intrusive load monitoring technology, integrating four-dimensional data: time, space, environment (comfort dimension), and behavior (economic dimension). By constructing a multi-dimensional energy consumption profile tag library, the system can accurately identify users' behavioral habits and environmental preferences, and generate personalized control strategies to improve user comfort and system energy efficiency.

[0056] After generating the image tag library, outdoor temperature is collected in real time, and based on the preset first image self-evolution rule, second image self-evolution rule and third image self-evolution rule, it is determined whether the image tag set of each region needs to be self-evolved. When it is determined, based on the first image self-evolution rule, that the image tag set for a certain region needs to undergo self-evolution, the temperature preference tag in the image tag set corresponding to that region is updated to generate a self-evolved image tag set. The first image self-evolution rule is determined when the frequency of user temperature adjustment in a certain region exceeds a set adjustment frequency threshold within a set second time period, indicating that the image tag set for that region needs to undergo self-evolution. Then, it continues to determine whether the image tag sets for each region need to undergo self-evolution based on the first, second, and third image self-evolution rules. Preferably, the second time period is one week, and the adjustment frequency threshold is 3 times / week.

[0057] When the frequency of user temperature adjustment in a certain area exceeds the set adjustment frequency threshold within the set second time period, it indicates that the user is frequently manually adjusting the temperature, indicating that the current system does not accurately grasp the comfort level. Therefore, the temperature preference label in the profile label set corresponding to that area needs to be updated. Updating the temperature preference tags in the image tag set corresponding to a certain area can be done by using steps 2 and 3.1, or by updating the temperature preference tags according to a method set by those skilled in the art. Preferably, the present invention updates the temperature preference tags based on the following method: Based on a preset temperature tolerance update formula, the temperature tolerance corresponding to the original temperature preference label is updated to obtain the updated temperature tolerance; the temperature tolerance update formula is as follows: Updated temperature tolerance = temperature tolerance decay coefficient × temperature tolerance corresponding to the original temperature preference label of the area + (1 - temperature tolerance decay coefficient) × min(user temperature adjustment frequency in the area ÷ adjustment frequency threshold, 1.0); The preferred value range for the temperature tolerance decay coefficient is 0.8; min(·) indicates taking the minimum value within the parentheses; Then, based on steps 2.2 and 3.1, the temperature preference labels in the region are updated according to the updated temperature tolerance in the region. Based on the updated temperature preference labels in the region, as well as the spatial activity labels, energy-active period labels, and electricity preference labels in the original image label set of the region, the original image label set of the region is updated to obtain the self-evolved image label set of the region.

[0058] When it is determined, based on the second image self-evolution rule, that the image tag set for a certain region needs to undergo self-evolution, the spatial activity tags in the corresponding image tag set for that region are updated to generate a self-evolved image tag set. The second image self-evolution rule determines that the image tag set for that region needs to undergo self-evolution when the rate of change of the spatial weight coefficient in a certain region exceeds a set threshold for the rate of change of the weight coefficient within a set third time period. Then, it continues to determine whether the image tag sets for each region need to undergo self-evolution based on the first, second, and third image self-evolution rules. Preferably, the third time period is one day, and the threshold for the rate of change of the weight coefficient is 20%. Those skilled in the art should know how to calculate the rate of change of the spatial weight coefficient, and it will not be elaborated here.

[0059] When the rate of change of spatial weight coefficient in a certain region within the set third time period is greater than the set threshold for the rate of change of weight coefficient, it indicates that the spatial usage pattern has changed significantly, indicating that the current spatial activity profile may be outdated. Therefore, the spatial activity tags in the profile tag set corresponding to that region need to be updated. Updating the spatial activity tags in the image tag set corresponding to a certain area can be done by updating the spatial activity tags using steps 2 and 3.1, or by updating the spatial activity tags according to a method set by those skilled in the art. Preferably, the present invention updates the spatial activity tags based on the following method: Based on a preset spatial weight coefficient update formula, the spatial weight coefficient corresponding to the original spatial activity tag is updated to obtain the updated spatial weight coefficient; the spatial weight coefficient update formula is as follows: Updated spatial weight coefficient = spatial weight decay coefficient × spatial weight coefficient corresponding to the original spatial activity label of the region + (1 - spatial weight decay coefficient) × min(the rate of change of spatial weight coefficient in the region ÷ the rate of change limit coefficient, 1.0); The preferred value range for the spatial weight attenuation coefficient is 0.8; min(·) represents taking the minimum value within the parentheses; the rate of change limit coefficient can be set by those skilled in the art, and its preferred value is 0.2. Then, based on steps 2.2 and 3.1, the spatial activity labels in the region are updated according to the updated spatial weight coefficients in the region. Based on the updated spatial activity labels in the region, as well as the energy-active period labels, temperature preference labels, and electricity preference labels in the original profile label set of the region, the original profile label set of the region is updated to obtain the self-evolved profile label set of the region.

[0060] When it is determined, based on the third self-evolution rule, that the image tag set for a certain region needs to undergo self-evolution, the energy-active period tags and electricity preference tags in the image tag set corresponding to that region are updated to generate a self-evolved image tag set. The third self-evolution rule determines that the image tag set for all regions needs to undergo self-evolution when the outdoor temperature change within a set fourth time period exceeds a set outdoor temperature change threshold. Then, it continues to determine whether the image tag set for each region needs to undergo self-evolution based on the first, second, and third self-evolution rules. Preferably, the fourth time period is 24 hours, and the outdoor temperature change threshold is 3-5°C. Those skilled in the art should know how to calculate the outdoor temperature change within the set fourth time period, which will not be elaborated here.

[0061] When the change in outdoor temperature exceeds the set outdoor temperature change threshold within the set fourth time period, it is determined that a sudden change in external temperature has occurred. The sudden change in temperature will change the user's energy consumption period pattern (time dimension) and sensitivity to electricity prices (economic dimension). Therefore, it is necessary to update the energy consumption active period label and electricity preference label in the profile label set corresponding to the region.

[0062] Updating the energy-active time period tags and electricity preference tags in the image tag set corresponding to a certain area can be done by using steps 2 and 3.1, or by updating the energy-active time period tags and electricity preference tags according to a method set by those skilled in the art. Preferably, the present invention updates the energy-active time period tags and electricity preference tags based on the following method: First, update the time period labels based on step 2.1; Secondly, based on a preset price sensitivity update formula, the price sensitivity corresponding to the original energy-active period label and electricity preference label is updated to obtain the updated price sensitivity; the price sensitivity update formula is as follows: Updated price sensitivity = price sensitivity attenuation coefficient × price sensitivity corresponding to the original active energy consumption period label and electricity preference label of the area + (1 - price sensitivity attenuation coefficient) × min(outdoor temperature change ÷ outdoor temperature change threshold, 1.0); The preferred value range for the price-sensitive attenuation coefficient is 0.8; min(·) indicates taking the minimum value within the parentheses; Then, based on steps 2.2 and 3.1, the energy-active time period labels and electricity preference labels in the region are updated based on the updated time period labels and price sensitivity in the region. Based on the updated energy-active time period labels and electricity preference labels in the region, as well as the spatial activity labels and temperature preference labels in the original profile label set of the region, the original profile label set of the region is updated to obtain the self-evolved profile label set of the region.

[0063] Add the self-evolved image tag sets of each region to the image tag library; If it is not determined based on the first, second, and third image self-evolution rules that the image tag set of any region needs to undergo self-evolution, then it is further determined based on the first, second, and third image self-evolution rules whether the image tag set of each region needs to undergo self-evolution.

[0064] This invention utilizes a self-evolutionary profile mechanism to enable the system to dynamically adjust extracted feature parameters (spatial weight coefficient, time period label, temperature tolerance, and price sensitivity) based on user feedback and environmental (temperature) changes. This allows for continuous optimization of distributed new energy air conditioning collaborative optimization instructions, thereby enhancing the system's intelligence level.

[0065] A collaborative optimization model is constructed; the collaborative optimization model includes a first scheduling layer for generating distributed new energy air conditioning collaborative optimization instructions within a preset optimization period, and a second scheduling layer for adjusting the collaborative optimization instructions in real time; The first scheduling layer includes a collaborative optimization objective function and collaborative optimization constraints; The collaborative optimization objective function aims to minimize the cost-comfort fusion data. The cost-comfort fusion data is calculated based on the amount of electricity purchased by the user from the grid, the electricity price, the comfort penalty weighting coefficient, the air conditioning set temperature of each region, and the user's comfort preference temperature in each region, as shown in the following formula: Cost-comfort fusion data = Total grid purchase cost + λ × Total comfort deviation penalty = Σt[ (P_grid,t × price_t) + λ × ΣiΣt[|T_set,i,t - T_pref,i,t| ] Where: Σt[·] represents summing all time period t data within the square brackets; P_grid,t represents the total electricity purchased by users from the grid during time period t; price_t represents the electricity price during time period t; ΣiΣt[·] represents summing all time period t data within the square brackets first, and then summing all data for region i within the square brackets; T_set,i,t represents the air conditioning set temperature for region i during time period t; T_pref,i,t represents the user's comfort preference temperature for region i during time period t; λ represents the comfort penalty weight coefficient, used to balance the system's economy and overall user comfort globally. The smaller the value of λ, the more emphasis is placed on cost, and the larger the value, the more emphasis is placed on comfort. The comfort penalty weight coefficient is determined based on the energy-saving preference label corresponding to the region and the preset correspondence between the energy-saving preference label and the comfort penalty weight coefficient. For example, for users with "energy-saving preference (κ=0.7)", λ may be lower (e.g., 0.3 yuan / ℃), focusing on economy; for users with "comfort priority (κ=0.1)", λ may be higher (e.g., 1.0 yuan / ℃). The user comfort preference temperature for the region can be calculated based on the user's preset baseline value T_base for the region and the temperature offset ΔT,t obtained based on the corresponding energy-active period label for the region and the preset correspondence between the energy-active period label and the temperature offset. For example, the correspondence between the energy-active period label and the temperature offset may be set to ΔT,t=0℃ for the region based on the "evening peak core" period label, and ΔT,t=−1℃ for the region based on the "night peak" period label.

[0066] The formula for calculating the user comfort preference temperature for the area is as follows: T_pref,i,t =T_base+ΔT,i,t, where: T_base is the user's global comfort reference temperature (e.g., 26℃), and ΔT,i,t is the temperature offset of the i-th region during time period t.

[0067] The first scheduling layer of this invention pre-generates distributed new energy air conditioning collaborative optimization instructions based on users' historical electricity consumption, profile tag library, electricity price, etc., so that the energy storage charging and discharging plan and the air conditioning operation strategy are coordinated and optimized, which can improve the local photovoltaic absorption rate and energy storage utilization efficiency, and achieve the technical effect of improving the overall energy utilization rate.

[0068] The collaborative optimization constraints include energy storage operation constraints, air conditioning power constraints, air conditioning temperature setting constraints, and power balance constraints. The power balance constraint is constructed based on the predicted photovoltaic power generation, energy storage output, the power purchased by users from the grid, and the total load demand of all regions, as shown in the following formula: Photovoltaic power generation during time period t + energy storage output + grid power purchase = total load demand for all regions.

[0069] The predicted photovoltaic power generation is obtained by inputting the photovoltaic power generation data from the historical user energy consumption data into a preset photovoltaic prediction model.

[0070] This invention drives the collaborative optimization model through a portrait tag library. The first scheduling layer of the collaborative optimization model enables the energy storage charging and discharging plan and the air conditioning operation strategy to be optimized in a coordinated manner, which can improve the local photovoltaic absorption rate and the energy storage utilization efficiency, thereby achieving the technical effect of improving the overall energy utilization rate.

[0071] The second scheduling layer calculates a power adjustment amount ΔP based on the room temperature and the air conditioner set temperature within a certain area when the absolute value of the deviation is greater than the temperature tolerance. The calculation is as follows: ΔP =K_p × K_c × (T_now - T_set) × H_s) Wherein, K_p is the power regulation ratio coefficient, with units of kW / ℃, and its preferred value range is 0.1 kW / ℃ to 0.5 kW / ℃. Those skilled in the art can fine-tune this range according to the air conditioning system specifications and building characteristics of the specific project. The constant temperature demand coefficient K_c = 1 / ΔT; ΔT represents the temperature tolerance; T_now represents the current temperature; T_set represents the current air conditioning set temperature; the spatial thermal inertia value H_s is preset by the equipment feature library, and the value varies for different spaces.

[0072] The frequency adjustment amount of the air conditioner compressor is determined based on the power adjustment amount; those skilled in the art should know how to determine the frequency adjustment amount of the air conditioner compressor based on the power adjustment amount, and will not be elaborated here.

[0073] The frequency of the air conditioner compressor is adjusted based on the aforementioned air conditioner compressor frequency adjustment amount.

[0074] The second scheduling layer of this invention sets out the principles and methods for real-time adjustment of distributed new energy air conditioning collaborative optimization instructions, so that the adjusted distributed new energy air conditioning collaborative optimization instructions are more in line with the current situation and continuously improve user satisfaction.

[0075] Before each optimization cycle, the profile tag library and electricity price are input into the first scheduling layer. Based on the solver, the collaborative optimization objective function and collaborative optimization constraints in the first scheduling layer are solved to generate the distributed new energy air conditioning collaborative optimization instruction for the optimization cycle. The distributed new energy air conditioning collaborative optimization command is executed, and the room temperature in each area is collected in real time and input into the second scheduling layer to adjust the distributed new energy air conditioning collaborative optimization command.

[0076] This invention utilizes a collaborative optimization model driven by a profile tag library, built upon a two-layer optimization control logic. This model enables collaborative optimization of energy storage charging and discharging plans and air conditioning operation strategies, thereby improving the local photovoltaic absorption rate and energy storage utilization efficiency, ultimately enhancing the overall energy utilization rate.

[0077] This invention has outstanding advantages such as forming a complete image tag library-driven control closed loop, being able to perform collaborative optimization of various devices, and generating collaborative optimization instructions that balance comfort and energy-saving requirements.

[0078] Specifically, when the power grid issues a demand response command, the average power purchased by users from the grid within a set fifth time period is collected as the first power purchase, and an area to be adjusted is selected from each of the aforementioned areas based on a preset area selection rule; the demand response command includes a power adjustment target value; preferably, the fifth time period is 1 minute. The area selection rule is that when the electricity price sensitivity of a certain area is higher than a preset third price sensitivity threshold and the spatial activity is lower than a preset activity threshold, that area is selected as the area to be adjusted; the spatial activity of each area is calculated based on the area dwell time and the number of people in the area. Preferably, the third price sensitivity threshold is 0.7 and the activity threshold is 50%.

[0079] This invention uses a regional selection rule to filter out non-critical regions for responding to demand response commands issued by the power grid, while also taking into account user comfort needs. This reduces electricity costs and enhances the stability and reliability of the power grid.

[0080] Based on the outdoor temperature and the room temperature of each of the areas to be adjusted, the maximum adjustable power of each of the areas to be adjusted is calculated accordingly. The maximum adjustable power of each region to be adjusted is the sum of the maximum adjustable power of all devices in the corresponding region under the current operating conditions.

[0081] The formula for calculating the maximum adjustable power of the equipment under the current operating conditions is as follows: ΔPmax = a × ΔTDR + b; in: ΔPmax is the maximum adjustable power of the device under the current operating conditions, measured in kilowatts (kW). It represents the upper limit of the adjustable power of the device without violating comfort constraints.

[0082] a and b are the first and second power-temperature response coefficients, respectively, which are model parameters obtained by linear regression fitting of historical operating data.

[0083] ΔTDR is the maximum permissible temperature adjustment range during a demand response event, expressed in degrees Celsius (°C). Its calculation formula is: ΔTDR=ΔT×(1−W s ); Where ΔT is the temperature tolerance of this area from the user profile tag library.

[0084] W s This is the spatial weighting coefficient for this region.

[0085] Based on the price sensitivity, dwell time, and number of residents in each of the regions to be adjusted, the regional adjustment priority of each of the regions to be adjusted is calculated. The formula for calculating the regional adjustment priority is as follows: Priority = κ×0.6 + (1-spatial activity)×0.4; Wherein, κ represents price sensitivity, and spatial activity is calculated based on the length of stay in the region and the number of people residing in the region.

[0086] Where t represents the t-th time a person stays in the area within the set first time period; T represents the total number of times a person stays in the area within the set first time period; when the area occupancy status of a certain area remains unchanged, but the number of people residing in the area changes, the area occupancy time is recalculated, and the number of times a person stays in the area is incremented by one; for example, if the area occupancy status of a certain area remains at 1, but the number of people residing in the area changes from 2 to 1, the number of times a person stays within the set first time period changes from 3 to 4. The preferred value range for the set first time period is 1 day.

[0087] The region to be adjusted corresponding to the maximum value among the adjustment priorities of each region is taken as the target adjustment region; the power of the target adjustment region is adjusted based on the maximum adjustable power of the target adjustment region; the adjustment must meet the temperature tolerance ΔT constraint.

[0088] After adjusting the power of the target adjustment area, the average power purchased by the user from the grid within a set fifth time period is collected as the second power purchase; preferably, the fifth time period is 1 minute.

[0089] When the deviation between the first purchased power and the second purchased power is greater than or equal to a set power threshold, and there is still an area to be adjusted, the area to be adjusted corresponding to the maximum value among the adjustment priorities of each area is taken as the target adjustment area; based on the maximum adjustable power of the target adjustment area, the power of the target adjustment area is adjusted until the deviation between the first purchased power and the second purchased power is less than the set power threshold, or there is no area to be adjusted; preferably, the power threshold is set based on the power adjustment target value, preferably set to 5% of the power adjustment target value.

[0090] The deviation ΔP between the first and second power purchases G The calculation formula is as follows: △P G= P T1 - P T2 ; Among them, P T1 The first power purchase capacity; P T2 This is the second power purchase capacity.

[0091] This invention can adjust the electricity consumption of non-critical areas according to the demand response instructions issued by the power grid, which can take into account both the user's comfort needs and the demand response instructions issued by the power grid, and achieve a win-win situation for both users and the power grid.

[0092] Example 1 A collaborative optimization method for distributed renewable energy air conditioning based on energy consumption profiles is provided in the appendix. Figure 1 .

[0093] Step 1: Collect historical user energy consumption data; the historical user energy consumption data includes photovoltaic power generation data, load data, non-intrusive load monitoring data, and environmental and human data.

[0094] In energy management systems that include various loads such as photovoltaic arrays, energy storage systems, and photovoltaic air conditioners (not limited to photovoltaic air conditioners, but also heat pumps, lighting systems, etc.), a sensing layer of devices is deployed, including photovoltaic monitoring equipment combinations (including photovoltaic inverter built-in monitoring modules, backplane temperature sensors, power meters, power quality analyzers, etc.), load monitoring equipment combinations (including current transformers, voltage sensors, etc.), environmental monitoring equipment combinations (including temperature and humidity sensors, CO2 sensors, human infrared thermal imaging sensors, etc.), and human body sensing devices (such as infrared sensors). These devices collect and upload user energy consumption-related data to the energy consumption profile processing platform in real time. The energy consumption profile processing platform generates optimization instructions and sends them to devices such as air conditioners and energy storage converters, forming a dual-loop coupling of energy flow and information flow.

[0095] The photovoltaic power generation data includes photovoltaic power generation, backsheet module temperature, photovoltaic power generation and voltage distortion rate collected by a combination of photovoltaic monitoring equipment; wherein the voltage distortion rate is collected by a power quality analyzer. The equipment operation data is collected by the load monitoring system, including the voltage, current, power, electricity consumption, and operating status of various electrical devices such as air conditioners. The operating status includes on-state, off-state, and fault state. Preferably, the operating status can be determined by comparing the voltage and current data detected in real time by the load monitoring device (such as a current transformer or voltage sensor) with the set voltage and current thresholds, respectively. Those skilled in the art should know how to set voltage and current thresholds to determine the operating status. Alternatively, the on-state and off-state can be distinguished by the presence or absence of voltage and current data, while the fault state is determined based on the results actively reported by the user.

[0096] Non-intrusive load monitoring: The non-intrusive load monitoring system (NILM technology) based on smart meters collects the current harmonics, active power, and reactive power at the main inlet of the user's electrical wire; as well as the power load of each independent electrical device (such as the power consumption of air conditioners, heat pumps, etc.). Environmental and Human Data: Temperature, humidity, CO2 concentration, and human infrared thermal signals are collected through an environmental monitoring system; the occupancy status (indicating whether someone is present in the area, 1 if present, 0 if absent), dwell time (indicating the length of time a person stays in the area), and number of people in the area are collected through human body sensing devices; when the occupancy status of an area remains unchanged, but the number of people in the area changes, the dwell time of that area is recalculated; in this invention, the preset areas refer to rooms such as living rooms and bedrooms in the building floor plan when not set by the user; if the user sets the division of areas in the building floor plan, the user's setting shall prevail.

[0097] Preferably, a weather station (such as an irradiance sensor) can be configured to monitor information such as irradiance.

[0098] Step 2: Based on the historical user energy consumption data, perform feature extraction to generate user energy consumption feature vectors for each preset area; Step 2.1: Based on the preset areas and the preset space-equipment association matrix in the building floor plan, extract features from the historical user energy consumption data to generate user energy consumption feature vectors for each preset area.

[0099] 2.1.1 Based on the pre-set space-equipment association matrix, clarify the correspondence between each area and each air conditioning unit in the building floor plan.

[0100] The space-device association matrix is ​​constructed based on the following method: The system's pre-set equipment feature library automatically matches areas (such as living rooms and bedrooms) in the building floor plan using equipment ID addresses, generating a space-equipment association matrix based on the electrical load of each independent electrical device. Since the physical location of each device ID within the building is recorded during installation, those skilled in the art can automatically match areas in the building floor plan using the equipment ID. The system's pre-set equipment feature library (including parameters such as air conditioner model, rated power, and thermal inertia coefficient of the area where the equipment is located) uses a dimensionless thermal inertia coefficient determined by the spatial material properties. Those skilled in the art should know how to obtain the thermal inertia coefficient of each area, which will not be elaborated here.

[0101] The space-device association matrix is ​​a static binary mapping between regions and device IDs (1 = associated, 0 = no association), and is only used to establish predefined relationships between space and devices.

[0102] In this matrix, rows correspond to areas defined on the building floor plan (such as living room, bedroom, kitchen, study, etc.); columns correspond to individual electrical appliances (such as air conditioner, heat pump, lighting system, etc.). The device ID is bound to the area during installation, so the column is identified by the device ID, not the device name; the values ​​in the matrix are 0 or 1, where 1 indicates that the device is installed in the area, that is, the device ID matches the area; 0 indicates that it is not installed in the area.

[0103] 2.1.2 Based on each preset area in the building floor plan, feature extraction is performed on the user energy consumption related data; the extracted features include spatial weight coefficient, time period label, temperature tolerance and price sensitivity.

[0104] The feature extraction of user energy consumption-related data based on each preset area in the building floor plan refers to extracting features of user energy consumption-related data in each preset area in terms of time dimension, spatial dimension, comfort dimension, and economic dimension, and obtaining time period label, spatial weight coefficient, temperature tolerance, and price sensitivity of each area.

[0105] In the spatial dimension, feature extraction is performed to obtain the spatial weight coefficients of each region. This is based on the regional dwell time and the number of people residing in each region, and the spatial weight coefficient W of each region is calculated. s The formula is as follows: W s = (Area of ​​the region × Spatial activity level of the region) ÷ Σ(Area of ​​all regions × Spatial activity level of the corresponding region); Wherein, Σ(area of ​​all regions × spatial activity of the corresponding region) means that after multiplying the area of ​​each region by its corresponding spatial activity, all product results are summed.

[0106] The spatial activity level of each area is calculated based on the duration of stay and the number of people residing in that area, as shown in the following formula: Spatial activity level of a certain area = total person-hours in that area ÷ sum of total person-hours in all areas; The formula for calculating the total per capita hours in the region is as follows: Total person-hours in the region = (Duration of stay in the area during the tth stay × Number of people in the area during the tth stay) Where t represents the t-th time a person stays in the area within the set first time period; T represents the total number of times a person stays in the area within the set first time period; when the area occupancy status of a certain area remains unchanged, but the number of people residing in the area changes, the area occupancy time is recalculated, and the number of times a person stays in the area is incremented by one; for example, if the area occupancy status of a certain area remains at 1, but the number of people residing in the area changes from 2 to 1, the number of times a person stays within the set first time period changes from 3 to 4. The preferred value range for the set first time period is 1 day.

[0107] Feature extraction in the time dimension to obtain time period labels involves clustering the historical electricity consumption data of each region using a preset clustering algorithm to determine the high-frequency energy consumption periods (e.g., 18:00-22:00) for each region. Based on these high-frequency energy consumption periods, time period labels (e.g., evening peak core, midday peak transition, evening peak segment, etc.) are generated for each region. Preferably, the preset clustering algorithm is a density-based noisy applied spatial clustering algorithm (DBSCAN clustering algorithm). Preferably, the preset clustering algorithm is a density-based noisy applied spatial clustering algorithm (DBSCAN clustering algorithm for short). Specifically, based on a preset clustering algorithm, the historical electricity consumption data of each region are clustered to determine the high-frequency energy consumption periods of each region, including: Based on historical electricity consumption data for each region, a time-electricity consumption correlation matrix is ​​constructed. The rows and columns of this matrix represent different initial time periods obtained by dividing 24 hours, and the columns and rows represent electricity consumption. In this matrix, the intersection of rows and columns records the average electricity consumption for the corresponding time period from the historical electricity consumption data used. Preferably, the different initial time periods obtained by dividing 24 hours are different initial time periods obtained by dividing 24 hours by an average, and the length of each initial time period is preferably in the range of 5 to 30 minutes. Preferably, one month's worth of historical electricity consumption data is used.

[0108] Based on a preset clustering algorithm, the historical electricity consumption data of each region are clustered to obtain multiple clusters; When the minimum power consumption of each initial period in the cluster is greater than or equal to the preset peak period power consumption baseline level, or greater than or equal to the first baseline level gain coefficient × the normal period power consumption baseline level, the period corresponding to the cluster is regarded as a high-frequency energy consumption period; the preferred value range of the first baseline level gain coefficient is 1.1~1.3, and this coefficient can be adjusted by those skilled in the art to ensure that at least one high-frequency energy consumption period is obtained.

[0109] The peak-hour electricity consumption baseline level and the normal-hour electricity consumption baseline level are obtained based on the following methods: The time-electricity consumption correlation matrix is ​​clustered using the k-means algorithm to obtain three clusters. The average electricity consumption of each of the three clusters is calculated, which is the cluster center. The maximum average electricity consumption is used as the peak-hour electricity consumption baseline, the median average electricity consumption is used as the normal-hour electricity consumption baseline, and the minimum average electricity consumption is used as the valley-hour electricity consumption baseline. Since multiple high-frequency energy consumption periods may be obtained, when generating period labels, those skilled in the art can select according to the actual situation. For example, a high-frequency energy consumption period can be randomly selected to generate a period label, or the period with the highest average electricity consumption among the high-frequency energy consumption periods can be selected to generate a period label.

[0110] The generation of time period tags for each region based on its high-frequency energy consumption periods is determined by the location of each region's high-frequency energy consumption periods within the day and the time range / average electricity consumption of the cluster corresponding to those periods. Specifically, the generation of time period tags for each region based on its high-frequency energy consumption periods includes: Based on the location of high-frequency energy consumption periods within a day, a first sub-label for each period is constructed. Further, when the high-frequency energy consumption period falls between 0:00 and 6:00 (inclusive of 0:00, exclusive of 6:00), the first sub-label is "Night"; when it falls between 6:00 and 12:00 (inclusive of 6:00, exclusive of 12:00), the first sub-label is "Morning"; when it falls between 12:00 and 18:00 (inclusive of 12:00, exclusive of 18:00), the first sub-label is "Afternoon"; and when it falls between 18:00 and 24:00 (inclusive of 18:00, exclusive of 24:00), the first sub-label is "Evening". Based on the time range of the cluster corresponding to this time period, or the average electricity consumption, a second sub-label for the time period is constructed, including: Determine whether the time range of the cluster corresponding to the time period is less than or equal to a set time period determination threshold; the preferred value of the time period determination threshold is 30 minutes. If the time range of the cluster corresponding to the time period is less than or equal to the set time period determination threshold, then the second sub-label of the time period is a transition; If the time range of the cluster corresponding to the time period is greater than the set time period determination threshold, then it is further determined whether the average power consumption of the cluster corresponding to the time period is greater than or equal to the product of the second baseline level gain coefficient and the peak time period baseline water; the preferred value range of the second baseline level gain coefficient is 1.2~1.5; the second baseline level gain coefficient is greater than or equal to the first baseline level gain coefficient.

[0111] If so, then the second sub-label of the time period is the core; If not, the second sub-label of the time period is empty.

[0112] When the second sub-label is not empty, the time period label is constructed according to the method of time period first sub-label + peak + time period second sub-label; for example, when the time period first sub-label is evening and the time period second sub-label is core, the time period label is evening peak core.

[0113] When the second sub-label is empty, the time period label is constructed according to the method of the first sub-label of the time period + the inter-peak segment; for example, when the first sub-label of the time period is evening and the second sub-label of the time period is empty, the time period label is the evening peak segment.

[0114] The time period labels exist in the form of numerical codes, for example, defining the evening peak core as 1, the morning peak core as 2, etc.

[0115] This invention can not only generate time period labels for each region corresponding to high-frequency energy consumption periods, but also generate time period labels for all time periods extracted based on clustering methods for each region, excluding high-frequency energy consumption periods. In other words, it generates time period labels for all time periods in each region, but only writes the time period labels corresponding to high-frequency energy consumption periods into the user energy consumption feature vector. The method for generating time period labels for all time periods in each region is as follows: The labels are first categorized based on the relative levels of electricity load, corresponding to the "peak, valley, and flat" electricity pricing system of the power grid (some regions use a system of peak, flat, valley, and peak periods; this invention applies this system to the user side). This categorization method is as follows: The criteria for dividing peak, flat, valley, and peak electricity consumption are as follows: Periods with electricity consumption 30% or less than the baseline level for normal or valley periods are considered low-consumption periods. Periods with electricity consumption 30% or more than the baseline level for peak periods are considered "high-frequency consumption periods." Periods with electricity consumption between 0.7 and 1.3 times the baseline level for normal periods are considered normal periods. Periods with electricity consumption exceeding 1.5 times the baseline level for peak periods are considered peak periods.

[0116] Based on the hierarchical classification, the time period is named according to its position in the day, making the labels more intuitive. For example, "morning," "noon," "evening," and "night." For the same type of time period, additional labels such as "core" and "transitional" can be added based on the stability and typicality of its electricity consumption patterns to refine the weighting.

[0117] Among them, "core" refers to the average electricity consumption of the clusters in the clustering results being more than 1.5 times the peak electricity consumption baseline level; "transition" refers to the period of short duration (such as the time range of the cluster being less than 30 minutes) and large fluctuation (such as fluctuations of more than 1.3 times the baseline level).

[0118] Possible tag examples: Evening peak core, Morning peak transition, Daytime flat period, Late night valley period, Midday peak (Here, peak refers to the period when the midday electricity consumption level is higher than the baseline peak electricity consumption level, and does not represent the peak, flat, valley, or peak mid-peak periods).

[0119] Temperature tolerance is obtained by feature extraction in the comfort dimension. This is based on user manual adjustment records and quantification of comfort preferences. Specifically, it is calculated based on the magnitude and number of times the user adjusts the temperature in each area within a preset first time period, as shown in the following formula: ΔT = Tolerance gain coefficient × Sum of temperature adjustments made by the user in this area within the first time period ÷ Number of temperature adjustments made by the user in this area within the first time period; Where ΔT represents temperature tolerance, the range by which the user adjusts the temperature in this area is the absolute value of the difference between the user's adjusted temperature in this area and the user's preset reference value. Preferably, the tolerance gain coefficient is 0.5, the first time length is 1 day, and when the user does not preset, the user's preset reference value is 26℃. For example, if the user's preset reference value is 26℃, the preferred value of the tolerance gain coefficient is 0.5, and the user manually adjusts the air conditioner 3 times in 1 day, adjusting it to 26.4℃, 26.6℃, and 25.8℃ respectively, with ranges of 0.4℃, 0.6℃, and 0.2℃ respectively, then ΔT = 0.5 × (0.4 + 0.6 + 0.2) ÷ 3 = 0.2℃; Price sensitivity is obtained by feature extraction in the economic dimension, based on electricity prices and the electricity consumption during the first and second electricity consumption periods in each region. According to the principles of adjacent first and second electricity consumption periods, different electricity prices during the first and second periods, and unchanged electricity prices within the first and second periods, those skilled in the art should know how to select the first and second electricity consumption periods; the formula for calculating price sensitivity is as follows: κ = [ (ΔQ / Q) / (ΔP / P) ] × e -0.05t ; Where κ represents price sensitivity (0≤κ≤1), e -0.05t The time decay factor is used, ΔQ / Q represents the relative rate of change in electricity consumption, ΔQ refers to the change in user electricity consumption between the first and second time periods, Q refers to the benchmark electricity consumption, ΔP / P represents the relative rate of change in electricity price, ΔP refers to the change in electricity price between the first and second time periods; P refers to the benchmark electricity price, e -0.05t The time decay factor is represented by t, which represents the interval between the user's most recent electricity price-sensitive behavior and the current time, in hours. 0.05 is the decay coefficient, in units of 1 / h. Those skilled in the art should know how to obtain the benchmark electricity consumption and benchmark electricity price, so they will not be elaborated here.

[0120] 2.2 Construct a user energy consumption feature vector for each region based on the spatial weight coefficient, the time period label, the temperature tolerance, and the price sensitivity.

[0121] The spatial weight coefficient W for each region obtained in step 2.1 will be used. s The time period label, temperature tolerance ΔT, and price sensitivity κ are combined into a vector form to serve as the user energy consumption feature vector for this region.

[0122] The user's energy consumption feature vector F is shown in the following formula: F = [spatial weighting coefficient, time period label, temperature tolerance, price sensitivity].

[0123] Step 3: Based on the preset profile tag generation rules, generate a profile tag library based on the user energy consumption feature vector of each region, and determine in real time whether to update the profile tag library.

[0124] 3.1 Based on the preset profile tag generation rules, a profile tag library is generated based on the user energy consumption feature vector of each region; the profile tag library includes the profile tag set of each region.

[0125] The image tag generation rules include sub-rules for generating spatial activity tags, sub-rules for generating energy-active time period tags, sub-rules for generating temperature preference tags, and sub-rules for generating electricity preference tags.

[0126] The spatial activity tag generation sub-rule compares the spatial weight coefficient of each region with a preset first spatial weight threshold and a second spatial weight threshold to determine the spatial activity level of each region. Based on preset region names (such as kitchen, living room, etc.) and the corresponding spatial activity level of the region, a spatial activity tag is generated. The first spatial weight threshold is less than the second spatial weight threshold. Specifically, when the spatial weight coefficient of a region is less than or equal to the first spatial weight threshold, the spatial activity level of that region is determined to be low. Preferably, the first spatial weight threshold is 0.5. When the spatial weight coefficient of a certain region is less than or equal to the second spatial weight threshold and greater than the first spatial weight threshold, the spatial activity level of the region is determined to be medium active; preferably, the second spatial weight threshold is 0.8. When the spatial weight coefficient of a certain region is greater than the second spatial weight threshold, the spatial activity level of that region is determined to be high activity.

[0127] The process of generating a space activity tag based on preset area names (such as kitchen, living room, etc.) and the corresponding space activity level of each area refers to merging the area name and the corresponding space activity level of the area, adding the word "area", and a space weight coefficient as the space activity tag. For example, when the area name is living room, the corresponding space activity level is high activity, and the space weight coefficient is 0.9, the generated space activity tag is "Living Room High Activity Area (W)". s =0.9).

[0128] The sub-rule for generating active energy consumption period labels is based on the average electricity consumption of the cluster corresponding to the high-frequency energy consumption period in each region and a preset baseline electricity consumption to calculate the active period weight. The baseline electricity consumption is preferably the average value of all historical electricity consumption data in the corresponding region.

[0129] Specifically, the active period weight is calculated based on the average power consumption of the clusters corresponding to the high-frequency energy consumption periods in a certain region and a preset baseline power consumption, including: Calculate the average power consumption of all data points within each cluster obtained by the clustering algorithm in step 2.1.2, and divide it by the baseline power consumption to obtain the initial relative activity index of each cluster.

[0130] The initial relative activity index of each cluster is normalized (e.g., by max-min normalization, Softmax function, etc.) and mapped to the range of (0,1) to obtain the corresponding normalized activity index.

[0131] The normalized activity index obtained based on the clusters corresponding to high-frequency energy consumption periods is used as the weight of the active energy consumption periods.

[0132] The price sensitivity of each region is compared with a preset first price sensitivity threshold and a second price sensitivity threshold to determine the price sensitivity level of each region. Based on the active time period weight, time period label and price sensitivity level of each region, an energy consumption active time period label is generated accordingly. The first price sensitivity threshold is greater than the second price sensitivity threshold. The price sensitivity of each region is compared with a preset first price sensitivity threshold and a second price sensitivity threshold to determine the price sensitivity level of each region, including: When κ ≥ a first price sensitivity threshold, the price sensitivity level is high; preferably, the first price sensitivity threshold is 0.7. When the second price sensitivity threshold ≤ κ < the first price sensitivity threshold, the price sensitivity level is medium sensitive; preferably, the second price sensitivity threshold is 0.4. When κ < the second price sensitivity threshold, the price sensitivity level is low.

[0133] The step of generating an energy-consuming active time period label based on the active time period weight, time period label, and price sensitivity level of each region refers to merging the first character of the time period label and the price sensitivity level of the region, and adding the active time period weight as the energy-consuming active time period label. For example, when the time period label is "evening peak core", the price sensitivity level is high sensitivity, and the active time period weight is 0.8, the energy-consuming active time period label is "evening peak high sensitivity (weight = 0.8)". The temperature preference label generation sub-rule compares the temperature tolerance of each region with a preset first temperature tolerance threshold and a second temperature tolerance threshold to determine the temperature preference of each region and generate a temperature preference label; the first temperature tolerance threshold is less than the second temperature tolerance threshold. The step of comparing the temperature tolerance of each region with a preset first temperature tolerance threshold and a second temperature tolerance threshold to determine the temperature preference of each region includes: When ΔT < the first temperature tolerance threshold, the temperature preference of this region is strictly isothermal; preferably, the first temperature tolerance threshold is 0.3℃. When the first temperature tolerance threshold ≤ ΔT < the second temperature tolerance threshold, the temperature preference of this region is mild and constant; preferably, the second temperature tolerance threshold is 0.8℃. When ΔT ≥ the second temperature tolerance threshold, the temperature preference in this region is relaxed and comfortable.

[0134] Temperature preference labels are generated based on temperature preference and temperature tolerance. For example, when the temperature preference is mild and constant temperature type and the temperature tolerance is 0.5, the temperature preference label is "mild and constant temperature type (ΔT=0.5℃)". The sub-rule for generating electricity preference tags is based on a preset price sensitivity level-electricity preference correspondence, generating electricity preference tags according to the price sensitivity level corresponding to each region. The price sensitivity level-electricity preference tag correspondence includes: High sensitivity (first price sensitivity threshold ≥ 0.7) corresponds to the electricity preference label "energy saving preference"; Medium sensitivity (second price sensitivity threshold ≤ κ < first price sensitivity threshold), corresponding to the electricity preference label "economic equilibrium type"; Low sensitivity (κ < second price sensitivity threshold) corresponds to the electricity preference label "comfort-first".

[0135] Electricity preference labels are generated based on the price sensitivity level and price sensitivity of each region. This involves determining electricity preferences based on the price sensitivity level and price sensitivity of each region, and generating electricity preference labels based on these preferences and price sensitivity. For example, when κ=0.7, the price sensitivity level is high sensitivity, the corresponding electricity preference is energy-saving preference, and the generated node preference label is "energy-saving preference (κ=0.7)". An initial profile tag set is generated based on the spatial activity tags, energy-active period tags, temperature preference tags, and electricity preference tags of each region.

[0136] The initial profile tag set for each region is set as follows: {"Spatial activity tag (spatial weight coefficient value)", "Energy active period tag (weight = energy active period weight)", "Temperature preference tag (ΔT = temperature tolerance)", "Electricity preference tag (κ = electricity price sensitivity)"}.

[0137] For example, when the living room's spatial activity level is labeled "High Activity Zone (W)" s =0.9), when the energy-intensive period is labeled "Evening Peak Sensitive Type (weight=0.8)", the temperature preference is labeled "Mild Constant Temperature Type (ΔT=0.5℃)", and the electricity preference is labeled "Energy Saving Preference (κ=0.7)", the initial profile label set for this area is: {"Living Room High Activity Area (W)"} s =0.9), “Evening Peak Sensitive Type (weight=0.8),” “Mild Constant Temperature Type (ΔT=0.5℃),” “Energy Saving Preference (κ=0.7)”}.

[0138] Use each initial image tag set as the current image tag set; 3.2 Determine in real time whether to update the image tag library.

[0139] After generating the image tag library in step 3.1, it is determined in real time whether to perform self-evolution of the image tag set for each region (referred to as image self-evolution). The profile self-evolution is based on certain triggering conditions. When any of the following events occur, the profile will initiate an update of relevant core parameters (spatial weight coefficient W_s, energy consumption active period weight, temperature tolerance ΔT, and price sensitivity factor κ) based on exponential smoothing to ensure that the profile dynamically adapts to changes in user behavior and the environment: The self-evolution of the image is based on certain triggering conditions. When the following events occur, the system independently updates the fusion weights of the specific dimensions most relevant to that event using exponential smoothing.

[0140] ① User temperature adjustment frequency > 3 times / week: Frequent manual temperature adjustment by users indicates that the current system does not accurately grasp the comfort level, so the comfort dimension weights should be updated; ② If the change rate of spatial weight coefficient is >20%, or the deviation of the usage rate of a certain area exceeds 20% (|current spatial activity of a certain area - historical average spatial activity| / historical average spatial activity >0.2): the spatial usage pattern has changed significantly, indicating that the current spatial activity profile may be outdated, and the spatial dimension weights should be updated; the historical average spatial activity is preferably calculated using historical data from the most recent 7 days (i.e., one week).

[0141] ③ Sudden change in external temperature > 5℃ / 24h: Sudden changes in temperature will change users' energy consumption patterns (time dimension) and sensitivity to electricity prices (economic dimension), so it is necessary to adjust the weights of these two dimensions.

[0142] Users adjust the temperature through mobile apps, web pages, desktop applications, and other terminal programs, and the frequency is higher than the specified value (e.g., users adjust the temperature more than 3 times / week). Compared with the historical average, the usage rate deviation in a certain area exceeds 20% (|current spatial activity of a certain area - historical average spatial activity| / historical average spatial activity > 0.2). The external temperature changes abruptly, and the change is greater than 5°C / 24h. The historical spatial activity average is preferably calculated using historical data from the most recent 7 days (i.e., one week).

[0143] The core parameter update formula is as follows: New parameter = α × Current parameter + (1-α) × Event trigger strength α = 0.8 (dynamic decay coefficient): Retains most historical states to ensure system stability. The tag library is the core input source for self-evolution, ensuring that profile updates are consistent with user's historical behavior.

[0144] For the "temperature adjustment frequency" event: Event trigger intensity = min(temperature adjustment frequency / adjustment frequency threshold, 1.0), for example, if the temperature is adjusted 4 times a week, the intensity = 4 / 3 ≈ 1.33, take 1.0, where the adjustment frequency threshold is preferably 3 times / week.

[0145] For the "spatial weight coefficient change" event: Event trigger intensity = min(spatial weight coefficient change rate in this area / 0.2, 1.0), for example, if the deviation is 25%, intensity = 0.25 / 0.2 = 1.25, take 1.0.

[0146] For the "Space Utilization Deviation" event: Event trigger intensity = min(relative deviation / 0.2, 1.0), for example, if the deviation is 25%, the intensity = 0.25 / 0.2 = 1.25, and we take 1.0.

[0147] For the "sudden change in external temperature" event: Event trigger intensity = min(outdoor temperature change / outdoor temperature change threshold, 1.0). For example, if the deviation is 25%, the intensity = 0.25 / 0.2 = 1.25, so we take 1.0.

[0148] The tag library is the core input source for self-evolution, ensuring that profile updates are consistent with users' historical behavior.

[0149] Step 4: Construct a collaborative optimization model; the collaborative optimization model includes a first scheduling layer for generating distributed new energy air conditioning collaborative optimization instructions within a preset optimization period, and a second scheduling layer for adjusting the collaborative optimization instructions in real time; The collaborative optimization model is constructed based on a two-layer optimization control logic, which consists of two layers: a day-ahead scheduling layer corresponding to the first scheduling layer and a real-time adjustment layer corresponding to the second scheduling layer.

[0150] The day-ahead scheduling layer, as the core of the upper-level decision-making, integrates input data such as photovoltaic power generation forecast data, grid time-of-use electricity price signals, and user profile tags to construct a mixed integer programming optimization model. The photovoltaic power generation prediction data is obtained by calling the meteorological data API of the 72 hours before the prediction or by the meteorological station configured in the system to obtain data such as irradiance, temperature, power generation, wind speed, wind direction, and air pressure. The photovoltaic power generation prediction model is then used to predict the photovoltaic power generation time series for the next 24 hours, thereby obtaining the photovoltaic power generation prediction data for each moment.

[0151] The photovoltaic power generation prediction model preferably uses an LSTM neural network.

[0152] The training set consists of a large number of samples. Each sample is constructed from 96 hours of historical data. The historical data for the first 72 hours includes meteorological data such as irradiance, temperature, power generation, wind speed, wind direction, and air pressure. The historical data for the last 24 hours is used as the photovoltaic power generation. The photovoltaic power generation of the last 24 hours is used as the label for the first 72 hours.

[0153] During training, the photovoltaic power generation prediction model uses meteorological data such as irradiance, temperature, power generation, wind speed, wind direction, and air pressure from the first 72 hours of each sample as input data; and photovoltaic power generation from the following 24 hours as output data. The mixed-integer programming optimization model aims to minimize the weighted sum of grid purchase cost and comfort deviation penalty (λ × |T_set - T_pref|, where λ represents the comfort penalty weight coefficient, λ=0.5 is the balance point between economy and comfort; T_set represents the user's preset value, such as a preset air conditioning temperature of 26℃; T_pref represents the user's preferred temperature). The weighted sum of grid purchase cost and comfort deviation penalty is denoted as cost-comfort fusion data. In other words, the mixed-integer programming optimization model aims to minimize the cost-comfort fusion data (also referred to as the objective function), and the cost-comfort fusion data is shown in the following formula: Cost-comfort fusion data = Total grid purchase cost + λ × Total comfort deviation penalty = Σt[ (P_grid,t × price_t) + λ × ΣiΣt[|T_set,i,t - T_pref,i,t| ] Where: Σt[·] represents summing all time period t data within the square brackets; P_grid,t represents the total electricity purchased by users from the grid during time period t; price_t represents the electricity price during time period t; ΣiΣt[·] represents summing all time period t data within the square brackets first, and then summing all data for region i within the square brackets; T_set,i,t represents the air conditioning set temperature for region i during time period t; T_pref,i,t represents the user's comfort preference temperature for region i during time period t; λ represents the comfort penalty weight coefficient, used to balance the system's economy and overall user comfort globally. The smaller the value of λ, the more emphasis is placed on cost, and the larger the value, the more emphasis is placed on comfort.

[0154] In the day-ahead scheduling layer, the division of time periods is based on the granularity requirements of the optimization model. In a specific implementation, the "day-ahead scheduling layer" generates a strategy for the next 24 time periods, preferably dividing the 24 hours into 24 time periods, each with a length of 1 hour. This division naturally aligns with photovoltaic power generation forecasts (outputting a 24-hour time series) and grid time-of-use pricing (usually in hourly units), facilitating model solving and strategy execution. Those skilled in the art can also adjust the time period length according to actual needs (e.g., 30 minutes), but a 1-hour time period is the preferred solution for balancing computational complexity and accuracy.

[0155] In the day-ahead scheduling layer's optimization model, T_set,i,t are decision variables pre-optimized by the model, rather than real-time adjusted values. The day-ahead scheduling layer calculates the planned set temperature for each time period in advance based on forecast data (such as photovoltaic power generation forecasts and electricity price signals) and user profiles (including comfort preferences).

[0156] T_pref,i,t are input parameters from the user profile, and their settings are based on the comfort dimension data in the profile tag library.

[0157] Basic settings: T_pref,i,t are usually user-preset baseline values ​​(such as 26℃). These values ​​are configured by the user or learned by the system based on historical data during profile initialization.

[0158] Furthermore, T_pref,i,t can be calculated based on the user-preset baseline value within the region and the temperature offset obtained based on the corresponding energy-active period label and the preset correspondence between the energy-active period label and the temperature offset. For example, based on the "evening peak core" period label, ΔT,t=0℃ may be set; based on the "night peak" period label, ΔT,t=−1℃ may be set.

[0159] For more refined settings, T_pref,i,t can be dynamically changed based on the time period labels for each time period within each region (time period labels for all time periods in each region need to be generated according to step 2.1.2). For example, T_pref,i,t might be set to 26℃ during the "evening peak core" period, and to 24℃ during the "late night valley" period.

[0160] The dynamically set formula is: T_pref,i,t = T_base + ΔT,i,t, where: T_base is the user-preset baseline value (e.g., 26℃), obtained from the "Comfort Dimension" in the profile tag library. ΔT,i,t is the temperature offset of the i-th region during time period t, determined according to the time period tag, such as: evening peak period (18:00-22:00): ΔT,t=0℃; Midnight Valley Period (22:00-06:00): Temperature shift ΔT,t=−2℃ during time period t.

[0161] The value of ΔT,t is learned by analyzing the correlation between historical user adjustment records and time period labels, and is specifically quantified using clustering or regression algorithms during user profile modeling. The quantification is based on the system calculating the average preferred temperature for each time period using historical data (such as user manual adjustment records). For example, the median or mean temperature ultimately set by users during the evening peak hours is used as the user's comfort preference temperature T_pref,t for time period t.

[0162] λ=0.5 serves as the equilibrium point, representing a penalty of 0.5 monetary units (e.g., yuan / ℃) for each unit of temperature deviation. The value of λ can be dynamically adjusted based on the "economic dimension" and "comfort dimension" tags in the user profile. For example, for users with a "power-saving preference (κ=0.7)," λ might be lower (e.g., 0.3 yuan / ℃), emphasizing economy; for "comfort-first" users, λ might be higher (e.g., 1.0 yuan / ℃), emphasizing comfort.

[0163] Simultaneously, collaborative optimization constraints are set, preferably, in this embodiment, they can also be called day-ahead scheduling constraints; the day-ahead scheduling constraints include energy storage operation constraints, air conditioning power constraints, air conditioning temperature setting constraints, and power balance constraints; Among them, the energy storage operation constraints strictly limit the energy storage SOC to the boundary range of 20% to 95%; Air conditioner power constraints: Ensure that the air conditioner power does not exceed 80% of its rated value. Air conditioning temperature setting constraints ensure that the indoor temperature is maintained within the comfort range of comfort preference ±ΔT; Power balance constraints ensure power balance (i.e., at any given moment, the photovoltaic power generation plus energy storage output equals the load demand plus the electricity purchased by the grid). The power balance constraints are constructed based on photovoltaic power generation, energy storage output, grid-purchased electricity, and total load demand across all regions. The objective function of the day-ahead dispatch layer is solved under the following power balance constraints: P PV,t + P bat,t +P grid,t = Σ i P load,i,t ; That is, the photovoltaic power generation P during time period t. PV,t + Energy storage output P bat,t + Power Grid Purchase P grid,t = Total load demand of all regions Σ i P load,i,t .

[0164] Before each optimization cycle (1 day in this embodiment), the profile tag library and the power grid time-of-use price signal are input into the first scheduling layer. Based on the solver, the collaborative optimization objective function and collaborative optimization constraints in the first scheduling layer are solved to generate the distributed new energy air conditioning collaborative optimization instruction for the optimization cycle. The solver efficiently calculates the future 24-hour timeframe and outputs a dynamic optimization strategy. For example, from 08:00 to 10:00, the energy storage charges 40kW, the air conditioner is set to 26℃, 30% of the load is supplied by photovoltaic power, and 70% is used for charging. From 12:00 to 14:00, the time is idle, the air conditioner is compensated to 24℃, and 100% of the load is supplied by photovoltaic power plus surplus electricity to the grid. From 18:30 to 20:30, the discharge is 35kW, the air conditioner is set to 25℃, and the power supply is mainly from energy storage supplemented by grid power. This achieves a refined dynamic balance between economy and user comfort. The solver can use methods such as branch and bound, and genetic algorithms for solving the problem.

[0165] Among them, the grid time-of-use price signal refers to obtaining the grid's published time-of-use price table for the next day, or directly reading the user-configured time-of-use price, which includes the time intervals for peak, valley, flat, and peak periods, as well as the corresponding price information for each period. Here, peak, valley, flat, and peak periods refer to the peak, valley, flat, and peak periods on the grid side, which are generally published by the grid.

[0166] User profile tag library: This refers to the results of energy consumption profiles, including tags for spatial activity, active energy consumption periods, temperature preferences, and electricity preferences, such as "high-activity area in the living room (W)". s =0.9), “Evening Peak Sensitive Type (weight=0.8),” “Mild Constant Temperature Type (ΔT=0.5℃),” “Energy Saving Preference (κ=0.7)”.

[0167] The strategy of the day-ahead scheduling layer is set one day in advance, for example, the optimization strategy for 00:00 to 24:00 the next day is generated at 23:00.

[0168] The hybrid integer programming model of the day-ahead scheduling layer of this invention can be replaced by a reinforcement learning model (such as DRL dynamic optimization).

[0169] Running the distributed new energy air conditioning collaborative optimization command, the real-time adjustment layer continuously reads the current room temperature T_now, the current optimized set temperature T_set, and the temperature tolerance ΔT data defined in the user profile. When the actual temperature deviates from the set value beyond the tolerance range (|T_now - T_set|>ΔT), a dynamic compensation mechanism based on proportional-integral control is immediately activated. This dynamic compensation mechanism calculates the precise power adjustment ΔP in real time using the constant temperature demand coefficient K_c and the spatial thermal inertia value H_s. ΔP = K_p × K_c × (T_now - T_set) × H_s) Wherein, K_p is the power regulation ratio coefficient, with units of kW / ℃, and its preferred value range is 0.1 kW / ℃ to 0.5 kW / ℃. Those skilled in the art can fine-tune this range according to the specifications of the air conditioning system and the building characteristics of the specific project. For example, for projects with high system inertia and high air conditioning power, a higher value (e.g., 0.4) can be used; for projects with low inertia and requiring precise control, a lower value (e.g., 0.15) can be used. The constant temperature demand coefficient K_c = 1 / ΔT; T_now represents the current temperature; T_set represents the current air conditioning set temperature; the spatial thermal inertia value H_s is preset by the equipment feature library, and the value varies for different spaces.

[0170] The air conditioner compressor frequency is finely adjusted based on the power adjustment amount (if ΔP ∈ [-50,0), the compressor frequency is lowered by 5~15Hz; if ΔP ∈ [0,50), the compressor frequency is increased by 5~15Hz). This, along with actions such as optimizing water pump flow, ensures the indoor temperature quickly returns to a comfortable range. In a variable frequency air conditioner control system, there is a clear mathematical relationship between power and frequency; those skilled in the art should know how to adjust the air conditioner compressor frequency based on the power adjustment amount. The preferred adjustment range for the compressor frequency is 25Hz~120Hz.

[0171] If the sensor detects no human activity for 5 consecutive minutes, it intelligently switches to energy-saving mode, slightly increasing the summer temperature setting or slightly decreasing it for winter to reduce energy consumption. Simultaneously, it automatically records user operation data every 5 minutes and determines whether to update the user profile label based on step 3.2. If the profile is updated, a real-time adjustment layer is triggered to make adjustments, creating an adaptive closed-loop optimization between comfort and energy efficiency, continuously improving user satisfaction and system operating efficiency. The fine-tuning amplitude is related to ΔT, and the preferred value range is as follows: Summer slight increase: Set the temperature increase range to (0.3 ~ 0.5) × ΔT.

[0172] Winter slight adjustment: The set temperature is adjusted down by (0.3 ~ 0.5) × ΔT.

[0173] If the value set in the adjustment command (such as reducing the compressor frequency or fan speed) is not reached within a specified time (e.g., 5 minutes) after the real-time adjustment layer issues an adjustment command, an alarm is triggered. Specifically, after the command is issued, parameter changes are continuously monitored, and an alarm is triggered if the expected trajectory is not reached within 5 minutes. For example, if the command is to reduce the compressor frequency by 15Hz, and the compressor frequency does not change within the specified time (e.g., 5 minutes) (the expectation here is that the compressor frequency decreases by 15Hz, or shows a downward trend but the decrease does not exceed 15Hz), an alarm is triggered.

[0174] When the power grid issues a demand response command, the decision engine is immediately activated. Based on the user profile tag library, non-critical areas with an electricity price sensitivity higher than 0.7 and a current spatial activity level of less than 50% are accurately selected as areas to be adjusted. Combining historical operating data (including outdoor temperature, room temperature of each area to be adjusted, and user power consumption), the maximum adjustable power of equipment in these spaces is predicted. The maximum adjustable power of equipment within these spaces is predicted by combining historical operating data, representing the upper limit of power adjustment without compromising comfort. The maximum adjustable power is dynamically calculated by fitting a power-temperature response model to the equipment based on historical operating data and considering the current spatial activity level. Specifically, firstly, historical operating data (including spatial activity levels, user adjustment records, and equipment power fluctuation data) for each space's equipment (such as air conditioners) is recorded in real time. Secondly, a space-equipment power response mapping model is established based on historical data. Finally, when the power grid triggers a DR command, the system calls upon non-critical areas with spatial activity levels <50% (such as unused study rooms) in real time, and calculates the maximum adjustable power using the model fitted from historical data for that area.

[0175] Specifically, the formula for calculating the maximum adjustable power is set as follows: When the power grid triggers a DR command, the maximum adjustable power ΔPmax of the internal equipment (taking air conditioning as an example) in each non-critical area with an activity level below 50% is calculated using the following model: ΔPmax = a × ΔTDR + b; in: ΔPmax is the maximum adjustable power of the device under the current operating conditions, measured in kilowatts (kW). It represents the upper limit of the adjustable power of the device without violating comfort constraints.

[0176] ΔTDR is the maximum permissible temperature adjustment range during a demand response event, expressed in degrees Celsius (°C). Its calculation formula is: ΔTDR=ΔT×(1−W s ); ΔT represents the temperature tolerance of this region from the user profile tag library.

[0177] W s This is the spatial weighting coefficient for this region.

[0178] a and b are the first and second power-temperature response coefficients, respectively, which are model parameters obtained by linear regression fitting of historical operating data.

[0179] Historical data refers to the historical operating data of the air conditioners in this area recorded by the system, including power P and the corresponding indoor-outdoor temperature difference ΔTreal (calculated as: outdoor temperature - indoor temperature).

[0180] The linear regression algorithm (such as the least squares method) was used to fit the relationship model P = a × ΔTreal + b.

[0181] Coefficient a (unit: kW / ℃) reflects the sensitivity of equipment power to temperature difference, while coefficient b (unit: kW) can be regarded as the base load.

[0182] While predicting the maximum adjustable power of equipment within these spaces by combining historical operational data, it is also necessary to dynamically generate adjustment strategies based on factors such as price sensitivity, spatial activity level, and temperature tolerance. The principle for generating these adjustment strategies is a triple balance principle, which prioritizes economic efficiency (price sensitivity), uses comfort (temperature tolerance) as a constraint boundary, and assigns dynamic adjustment weights based on spatial importance (spatial activity level), aiming to "minimize comfort loss and maximize economic benefits." Specifically, price sensitivity κ determines the priority of economic benefits, spatial activity level determines the adjustable space range, and temperature tolerance ΔT constrains the adjustment range.

[0183] First, calculate the regional adjustment priority (hereinafter referred to as priority) for each non-critical region, as shown in the following formula: Priority = κ×0.6 + (1-spatial activity)×0.4; For high-priority areas, adjustments are made first (e.g., air conditioning temperatures are raised first in areas with high κ). The adjustment range is strictly limited by ΔT. Areas with low activity can withstand larger adjustment ranges (e.g., raising the temperature by 1.5℃), while areas with high activity only undergo minor adjustments (e.g., raising the temperature by 0.5℃). Load reduction data is transmitted back in real time through smart meters, and the actual reduction amount is verified in a closed loop to check the deviation from the grid target.

[0184] The generated regulation strategies, such as raising the set temperature of the air conditioner in a selected space by 1.5°C in summer, suspending the operation of the pool water pump, and increasing the energy storage discharge power, are implemented. During execution, load reduction data is transmitted in real time through smart meters, forming a demand response command issued by the power grid. The system executes the strategy, the smart meters monitor the actual adjustment amount, and the system performs closed-loop verification with a deviation of ≤5%, ensuring that the actual reduction amount deviates strictly from the power grid target by less than 5%. This ensures efficient response to power grid dispatching needs while guaranteeing user comfort, achieving a dynamic balance between economy and reliability. The power grid regulation target is determined based on the standard demand response command issued by the power grid. Specifically, the power grid sends DR commands to the user system through the standard power communication protocol, and the user system parses the commands to obtain the power grid target.

[0185] In terms of fault handling and safety assurance, the present invention ensures stable operation and user privacy through intelligent redundancy mechanism and multiple safety interlock design. Specifically, the multiple safety interlock design includes over-temperature protection and privacy protection.

[0186] Among them, the over-temperature protection means that when the return air temperature is greater than 40°C, the compressor will be forcibly shut down and the cooling fan will be started. Privacy protection means processing human movement data locally. Specifically, this involves removing personal identification information such as IDs and trajectory coordinates locally to complete the desensitization process, and only uploading the spatial activity of each region.

[0187] The intelligent redundancy mechanism automatically switches to the standby mode of the non-intrusive load monitoring (NILM) function when the confidence level of the NILM decomposition falls below 85%. The standby mode utilizes the total power data of the smart meter combined with historical average load curves of individual devices, and predicts the power load of each independent electrical device (such as the power consumption of air conditioners and heat pumps) using a BP neural network. Simultaneously, it sends real-time alarm prompts to calibrate sensors such as the smart meter. The input data for the BP neural network includes the total power data of the smart meter and historical average load curves of individual devices. The data used covers the most recent 7 days. The NILM of this invention can be replaced by smart meter + AI-based sub-metering (e.g., predicting sub-item loads by training an LSTM network using the current waveform characteristics of the meter).

[0188] The safety interlock protects critical components in real time—once the return air temperature exceeds 40°C, the system immediately forces the compressor to shut down and starts the cooling fan to prevent the equipment from overheating. When the grid voltage is higher than 265V, the system automatically disconnects the grid-connected contactor to prevent energy storage backflow. At the same time, all human movement data is processed locally, and only the desensitized activity index is uploaded to protect privacy. This ensures reliable system operation in the event of a fault, balancing safety protection and user data security.

[0189] Those skilled in the art will understand that the specific implementation methods of photovoltaic air conditioning, NILM monitoring, and mixed integer programming described above are merely examples, and the present invention can be applied to photovoltaic heat pump systems, smart meter sub-metering, reinforcement learning optimization, and other aspects.

[0190] When this invention is applied to a photovoltaic heat pump system, the photovoltaic air conditioner can be replaced by a photovoltaic heat pump system, or a photovoltaic-energy storage-load integrated device (such as a cabinet terminal that integrates photovoltaic panels, energy storage modules, and air conditioning compressors).

[0191] The second aspect of this application discloses a distributed new energy air conditioning collaborative optimization system based on a distributed new energy air conditioning collaborative optimization method, including a user energy consumption data acquisition module, a user energy consumption feature vector generation module, a profile tag library generation module, a collaborative optimization model construction module, a collaborative optimization instruction generation module, and a collaborative optimization instruction real-time adjustment module: The user energy consumption data acquisition module is used to collect historical user energy consumption data; The user energy consumption feature vector generation module performs feature extraction based on the historical user energy consumption related data to generate user energy consumption feature vectors for each preset area. The image tag library generation module generates an image tag library based on the user energy consumption feature vector of each region according to the preset image tag generation rules. The collaborative optimization model construction module is used to construct a collaborative optimization model; the collaborative optimization model includes a first scheduling layer for generating distributed new energy air conditioning collaborative optimization instructions within a preset optimization period, and a second scheduling layer for adjusting the collaborative optimization instructions in real time. The collaborative optimization instruction generation module is used to input the profile tag library and electricity price into the first scheduling layer before each optimization cycle, solve the collaborative optimization objective function and collaborative optimization constraints in the first scheduling layer based on the solver, and generate the distributed new energy air conditioning collaborative optimization instruction for the optimization cycle. The real-time adjustment module for the collaborative optimization instruction is used to run the distributed new energy air conditioning collaborative optimization instruction, and to collect the room temperature input of each area in real time into the second scheduling layer to adjust the distributed new energy air conditioning collaborative optimization instruction.

[0192] This invention can also incorporate a local profile processing module based on edge computing. Deployed in the home energy gateway, this module enables localized processing and storage of user profile data, avoiding the uploading of sensitive behavioral data to the cloud and enhancing privacy protection. Simultaneously, this module supports parallel processing of multiple user profiles, making it suitable for complex scenarios with different family members and varying lifestyles. By comparing the priorities of each member's profile, the system makes intelligent decisions under conflict settings, such as prioritizing the comfort needs of the elderly or infants, further enhancing the system's intelligence and user-friendliness.

[0193] The priority division of the member profiles is based on the temperature tolerance ΔT by default. The smaller the ΔT, the higher the priority; or it can be based on the priority set by the user through the software.

[0194] The conflict setting refers to the situation where the comfort needs of multiple users cannot be met simultaneously under physical constraints. For example, insufficient power supply from photovoltaic / energy storage may prevent the comfort settings of all users from being met at the same time (e.g., the elderly need 26°C, while young people need 24°C).

[0195] The intelligent decision-making under conflict conditions is achieved through a priority-weighted dynamic power allocation algorithm, directly adopting the day-ahead scheduling layer and real-time adjustment layer logic from step 4, only extending the multi-user dimension. Specifically, firstly, the priority values ​​of each member profile are input (determined based on the temperature tolerance ΔT calculation result in step 2.2), current photovoltaic power, energy storage SOC, grid electricity price, real-time temperature, and other real-time data, along with conflict constraints (such as the upper limit of total air conditioning power). Then, the overall comfort target is calculated through priority weighting. Under the constraint of the upper limit of total air conditioning power, the solution is: min(grid electricity purchase cost + Σi(priorityᵢ × comfort deviation)), and the real-time adjustment layer dynamically adjusts.

[0196] See appendix Figure 2The user energy consumption data acquisition module is located in the perception layer to collect data, including the total incoming current waveform collected by the current sensor, room temperature, humidity, and human presence collected by the environmental sensor, and inverter output power generation data collected by the photovoltaic monitoring terminal. Data flow transmission occurs between the perception layer data acquisition module and the current sensor, environmental sensor, and photovoltaic monitoring terminal. The user energy consumption feature vector generation module and the profile tag library generation module are located in the profile server, which is connected to the intelligent distribution cabinet and transmits the generated profile tag set to the intelligent distribution cabinet. Data flow transmission exists between the profile server and the intelligent distribution cabinet. The collaborative optimization model construction module and the collaborative optimization instruction generation module are located in the intelligent distribution cabinet. Based on the distributed new energy air conditioning collaborative optimization instructions generated by the collaborative optimization instruction generation module, the input / output power of the energy storage system, air conditioning system, and photovoltaic array is adjusted. Energy flow transmission occurs between the intelligent distribution cabinet and the energy storage system, air conditioning system, and photovoltaic array. The collaborative optimization instruction real-time adjustment module is set in the decision server, which reads the data collected from the perception layer in real time (there is a data stream transmission between the two), and transmits the adjustment results to the air conditioning controller through the data stream. The air conditioning controller controls the compressor / fan execution terminal to make real-time adjustments.

[0197] The distributed new energy air conditioning collaborative optimization system of the present invention achieves collaborative energy management through a closed-loop chain: a user energy consumption data acquisition module for data acquisition, a user energy consumption feature vector generation module for feature extraction, a profile tag library generation module for generating energy consumption profiles, a collaborative optimization instruction generation module for generating energy consumption strategies, and a collaborative optimization instruction real-time adjustment module for executing energy consumption strategies and continuously optimizing.

[0198] The third aspect of the present invention discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the distributed new energy air conditioning collaborative optimization method.

[0199] This invention can be applied to any computer program and has a wide range of applications.

[0200] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0201] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0202] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0203] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0204] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A distributed new energy air conditioner collaborative optimization method based on energy portrait, characterized in that, The method comprises: collecting historical user energy consumption related data; performing feature extraction based on the historical user energy consumption related data to generate user energy consumption feature vectors of each preset region; generating a portrait label library based on the user energy consumption feature vectors of each region according to a preset portrait label generation rule; constructing a collaborative optimization model, wherein the collaborative optimization model comprises a first scheduling layer for generating distributed new energy air conditioner collaborative optimization instructions within a preset optimization period, and a second scheduling layer for adjusting the collaborative optimization instructions in real time; inputting the portrait label library and electricity price into the first scheduling layer before each optimization period to generate the distributed new energy air conditioner collaborative optimization instructions within the optimization period; running the distributed new energy air conditioner collaborative optimization instructions and collecting room temperature in each region in real time to adjust the distributed new energy air conditioner collaborative optimization instructions.

2. The distributed new energy air conditioner collaborative optimization method according to claim 1, wherein: the historical user energy consumption related data comprises power consumption, region residence duration and region residence population of each device in each region.

3. The distributed new energy air conditioner collaborative optimization method according to claim 1, wherein: the user energy consumption feature vector of each region comprises a space weight coefficient, a time period label, a temperature tolerance and a price sensitivity.

4. The distributed new energy air conditioner collaborative optimization method according to claim 1 or 2 or 3, wherein: the feature extraction based on the historical user energy consumption related data to generate the user energy consumption feature vectors of each preset region comprises: performing feature extraction based on the historical user energy consumption related data to obtain the space weight coefficient, the time period label, the temperature tolerance and the price sensitivity of each region; constructing the user energy consumption feature vector of each region based on the space weight coefficient, the time period label, the temperature tolerance and the price sensitivity of each region.

5. The distributed new energy air conditioner collaborative optimization method according to claim 4, wherein: the feature extraction based on the historical user energy consumption related data to obtain the space weight coefficient, the time period label, the temperature tolerance and the price sensitivity of each region comprises: calculating the space weight coefficient of each region based on the region area, the region residence duration and the region residence population of each region; determining the high-frequency energy consumption time period of each region by clustering the historical power consumption data of each region based on a preset clustering algorithm, and generating the time period label of each region based on the high-frequency energy consumption time period of each region; correspondingly calculating the temperature tolerance of each region based on the amplitude and frequency of temperature adjustment of users in each region within a preset first time length; correspondingly calculating the price sensitivity of each region based on the electricity price and the power consumption of each region.

6. The distributed new energy air conditioner collaborative optimization method according to claim 1, wherein: The image label library is generated based on the user energy feature vectors of the regions according to a preset image label generation rule, and the image label library comprises: The image label library is generated based on the user energy feature vectors of the regions according to a preset image label generation rule, and the image label library comprises: The image label library is generated based on the user energy feature vectors of the regions according to a preset image label generation rule, and the image label library comprises: The image label library is generated based on the user energy feature vectors of the regions according to a preset image label generation rule, and the image label library comprises:

7. The distributed new energy air conditioner collaborative optimization method according to claim 1 or 5 or 6, characterized in that: The image label generation rule comprises a space activity level label generation sub-rule, an energy active period label generation sub-rule, a temperature preference label generation sub-rule, and an electricity preference label generation sub-rule; The space activity level label generation sub-rule compares the space weight coefficient of each region with a preset first space weight threshold and a second space weight threshold, and determines the space activity level of each region correspondingly; The space activity level label is generated based on the region name and the space activity level of the corresponding region; The first space weight threshold is smaller than the second space weight threshold; The energy active period label generation sub-rule calculates an active period weight based on the average electricity consumption of the cluster corresponding to the high-frequency energy consumption period of each region and a preset baseline electricity consumption, compares the price sensitivity of each region with a preset first price sensitivity threshold and a second price sensitivity threshold, and determines the price sensitivity level of each region correspondingly; The energy active period label is generated based on the active period weight, the period label, and the price sensitivity level of each region; The first price sensitivity threshold is greater than the second price sensitivity threshold; The temperature preference label generation sub-rule compares the temperature tolerance of each region with a preset first temperature tolerance threshold and a second temperature tolerance threshold, and determines the temperature preference label of each region correspondingly; The first temperature tolerance threshold is smaller than the second temperature tolerance threshold; The electricity preference label generation sub-rule generates the electricity preference label of each region based on the price sensitivity level according to a preset price sensitivity level-electricity preference correspondence.

8. The distributed new energy air conditioner collaborative optimization method according to claim 1 or 6, characterized in that: After the image label library is generated, whether the image label set of each region needs to be self-evolved is determined based on a preset first image self-evolution rule, a second image self-evolution rule, and a third image self-evolution rule respectively. When it is determined, based on the first image self-evolution rule, that the image label set of a certain region needs to be self-evolved, the temperature preference label in the image label set of the region is updated to generate a self-evolved image label set; the first image self-evolution rule is that when the frequency of user adjustment of temperature in a certain region is greater than a set adjustment frequency threshold within a set second time length, it is determined that the image label set of the region needs to be self-evolved; When it is determined, based on the second image self-evolution rule, that the image label set of a certain region needs to be self-evolved, the space activity label in the image label set of the region is updated to generate a self-evolved image label set; the second image self-evolution rule is that when the rate of change of the space weight coefficient in a certain region is greater than a set weight coefficient change rate threshold within a set third time length, it is determined that the image label set of the region needs to be self-evolved; When it is determined, based on the third image self-evolution rule, that the image label set of a certain region needs to be self-evolved, the energy use active period label and the electricity use preference label in the image label set of the region are updated to generate a self-evolved image label set; the third image self-evolution rule is that when the amount of change of outdoor temperature is greater than a set outdoor temperature change threshold within a set fourth time length, it is determined that the image label set of all regions needs to be self-evolved; If it is not determined, based on the first image self-evolution rule, the second image self-evolution rule and the third image self-evolution rule, that the image label set of any region needs to be self-evolved, then it is determined, based on the first image self-evolution rule, the second image self-evolution rule and the third image self-evolution rule, whether the image label set of each region needs to be self-evolved.

9. The distributed new energy air conditioner collaborative optimization method according to claim 1, characterized in that: the first scheduling layer comprises a collaborative optimization objective function and a collaborative optimization constraint; the collaborative optimization objective function takes minimizing cost comfort fusion data as a goal; the collaborative optimization constraint comprises energy storage operation constraint, air conditioner power constraint, air conditioner temperature setting constraint and power balance constraint.

10. The distributed new energy air conditioner collaborative optimization method according to claim 6 or 9, characterized in that: the cost comfort fusion data is obtained based on user electricity purchase amount, electricity price, comfort penalty weight coefficient, air conditioner set temperature of each region and user comfort preference temperature of each region; the comfort penalty weight coefficient is determined based on the electricity saving preference label corresponding to the region and a preset electricity saving preference label-comfort penalty weight coefficient correspondence; the user comfort preference temperature of the region is obtained based on a preset reference value of the user in the region and a temperature offset obtained based on a preset energy use active period label-temperature offset correspondence.

11. The distributed new energy air conditioner collaborative optimization method according to claim 1 or 5 or 6, characterized in that: The second scheduling layer, when the absolute value of the deviation of the room temperature from the air conditioner set temperature in a certain region is greater than the temperature tolerance, calculates a power adjustment amount based on the room temperature and the air conditioner set temperature in the region; determining an air conditioner compressor frequency adjustment amount based on the power adjustment amount; adjusting the air conditioner compressor frequency based on the air conditioner compressor frequency adjustment amount.

12. The distributed new energy air conditioner collaborative optimization method according to claim 1 or 5, characterized in that: when the power grid issues a demand response instruction, collect the average power purchased by the user from the power grid within a set fifth time length as a first power purchase, and select a region to be adjusted in each of the regions based on a preset region selection rule; based on the outdoor temperature and the room temperature of each of the regions to be adjusted, calculate the maximum adjustable power of each of the regions to be adjusted; based on the price sensitivity, the region residence time and the number of people in each of the regions to be adjusted, calculate the region adjustment priority of each of the regions to be adjusted; select the region to be adjusted corresponding to the maximum value of the region adjustment priority as a target adjustment region; adjust the power of the target adjustment region based on the maximum adjustable power of the target adjustment region; after adjusting the power of the target adjustment region, collect the average power purchased by the user from the power grid within a set fifth time length as a second power purchase; when the deviation between the first power purchase and the second power purchase is greater than or equal to a set power threshold, and there is still the region to be adjusted, continue to select the region to be adjusted corresponding to the maximum value of the region adjustment priority as a target adjustment region, and adjust the power of the target adjustment region based on the maximum adjustable power of the target adjustment region, until the deviation between the first power purchase and the second power purchase is less than the set power threshold, or there is no region to be adjusted.

13. The distributed new energy air conditioner collaborative optimization method according to claim 12, characterized in that: the region selection rule is that when the price sensitivity of a certain region is higher than a preset third price sensitivity threshold and the space activity is lower than a preset activity threshold, the region is selected as the region to be adjusted, and the space activity of each region is calculated based on the region residence time and the number of people in the region.

14. A distributed new energy air conditioner collaborative optimization system using the distributed new energy air conditioner collaborative optimization method of any one of claims 1-13, characterized in that, including a user energy consumption data acquisition module, a user energy consumption feature vector generation module, a portrait label library generation module, a collaborative optimization model construction module, a collaborative optimization instruction generation module and a real-time collaborative optimization instruction adjustment module: the user energy consumption data acquisition module is used to acquire historical user energy consumption related data; the user energy consumption feature vector generation module extracts features based on the historical user energy consumption related data to generate user energy consumption feature vectors of each preset region; the portrait label library generation module generates a portrait label library based on the user energy consumption feature vectors of each region according to a preset portrait label generation rule; The collaborative optimization model construction module is used to construct a collaborative optimization model; the collaborative optimization model includes a first scheduling layer for generating distributed new energy air conditioning collaborative optimization instructions within a preset optimization period, and a second scheduling layer for adjusting the collaborative optimization instructions in real time. The collaborative optimization instruction generation module is used to input the profile tag library and electricity price into the first scheduling layer before each optimization cycle to generate the distributed new energy air conditioning collaborative optimization instruction for that optimization cycle. The real-time adjustment module for the collaborative optimization instruction is used to run the distributed new energy air conditioning collaborative optimization instruction, and to collect the room temperature input of each area in real time into the second scheduling layer to adjust the distributed new energy air conditioning collaborative optimization instruction.

15. A computer readable storage medium having stored thereon a computer program, characterized in that, When executed by the processor, the program implements the steps of the distributed new energy air conditioning collaborative optimization method according to any one of claims 1-13.

Images